2 * Kernel-based Virtual Machine driver for Linux
4 * This module enables machines with Intel VT-x extensions to run virtual
5 * machines without emulation or binary translation.
7 * Copyright (C) 2006 Qumranet, Inc.
8 * Copyright 2010 Red Hat, Inc. and/or its affiliates.
11 * Avi Kivity <avi@qumranet.com>
12 * Yaniv Kamay <yaniv@qumranet.com>
14 * This work is licensed under the terms of the GNU GPL, version 2. See
15 * the COPYING file in the top-level directory.
19 #include <kvm/iodev.h>
21 #include <linux/kvm_host.h>
22 #include <linux/kvm.h>
23 #include <linux/module.h>
24 #include <linux/errno.h>
25 #include <linux/percpu.h>
27 #include <linux/miscdevice.h>
28 #include <linux/vmalloc.h>
29 #include <linux/reboot.h>
30 #include <linux/debugfs.h>
31 #include <linux/highmem.h>
32 #include <linux/file.h>
33 #include <linux/syscore_ops.h>
34 #include <linux/cpu.h>
35 #include <linux/sched.h>
36 #include <linux/cpumask.h>
37 #include <linux/smp.h>
38 #include <linux/anon_inodes.h>
39 #include <linux/profile.h>
40 #include <linux/kvm_para.h>
41 #include <linux/pagemap.h>
42 #include <linux/mman.h>
43 #include <linux/swap.h>
44 #include <linux/bitops.h>
45 #include <linux/spinlock.h>
46 #include <linux/compat.h>
47 #include <linux/srcu.h>
48 #include <linux/hugetlb.h>
49 #include <linux/slab.h>
50 #include <linux/sort.h>
51 #include <linux/bsearch.h>
53 #include <asm/processor.h>
55 #include <asm/ioctl.h>
56 #include <asm/uaccess.h>
57 #include <asm/pgtable.h>
59 #include "coalesced_mmio.h"
63 #define CREATE_TRACE_POINTS
64 #include <trace/events/kvm.h>
66 /* Worst case buffer size needed for holding an integer. */
67 #define ITOA_MAX_LEN 12
69 MODULE_AUTHOR("Qumranet");
70 MODULE_LICENSE("GPL");
72 /* Architectures should define their poll value according to the halt latency */
73 static unsigned int halt_poll_ns
= KVM_HALT_POLL_NS_DEFAULT
;
74 module_param(halt_poll_ns
, uint
, S_IRUGO
| S_IWUSR
);
76 /* Default doubles per-vcpu halt_poll_ns. */
77 static unsigned int halt_poll_ns_grow
= 2;
78 module_param(halt_poll_ns_grow
, uint
, S_IRUGO
| S_IWUSR
);
80 /* Default resets per-vcpu halt_poll_ns . */
81 static unsigned int halt_poll_ns_shrink
;
82 module_param(halt_poll_ns_shrink
, uint
, S_IRUGO
| S_IWUSR
);
87 * kvm->lock --> kvm->slots_lock --> kvm->irq_lock
90 DEFINE_SPINLOCK(kvm_lock
);
91 static DEFINE_RAW_SPINLOCK(kvm_count_lock
);
94 static cpumask_var_t cpus_hardware_enabled
;
95 static int kvm_usage_count
;
96 static atomic_t hardware_enable_failed
;
98 struct kmem_cache
*kvm_vcpu_cache
;
99 EXPORT_SYMBOL_GPL(kvm_vcpu_cache
);
101 static __read_mostly
struct preempt_ops kvm_preempt_ops
;
103 struct dentry
*kvm_debugfs_dir
;
104 EXPORT_SYMBOL_GPL(kvm_debugfs_dir
);
106 static int kvm_debugfs_num_entries
;
107 static const struct file_operations
*stat_fops_per_vm
[];
109 static long kvm_vcpu_ioctl(struct file
*file
, unsigned int ioctl
,
111 #ifdef CONFIG_KVM_COMPAT
112 static long kvm_vcpu_compat_ioctl(struct file
*file
, unsigned int ioctl
,
115 static int hardware_enable_all(void);
116 static void hardware_disable_all(void);
118 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
);
120 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
);
121 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
, gfn_t gfn
);
123 __visible
bool kvm_rebooting
;
124 EXPORT_SYMBOL_GPL(kvm_rebooting
);
126 static bool largepages_enabled
= true;
128 bool kvm_is_reserved_pfn(kvm_pfn_t pfn
)
131 return PageReserved(pfn_to_page(pfn
));
137 * Switches to specified vcpu, until a matching vcpu_put()
139 int vcpu_load(struct kvm_vcpu
*vcpu
)
143 if (mutex_lock_killable(&vcpu
->mutex
))
146 preempt_notifier_register(&vcpu
->preempt_notifier
);
147 kvm_arch_vcpu_load(vcpu
, cpu
);
151 EXPORT_SYMBOL_GPL(vcpu_load
);
153 void vcpu_put(struct kvm_vcpu
*vcpu
)
156 kvm_arch_vcpu_put(vcpu
);
157 preempt_notifier_unregister(&vcpu
->preempt_notifier
);
159 mutex_unlock(&vcpu
->mutex
);
161 EXPORT_SYMBOL_GPL(vcpu_put
);
163 static void ack_flush(void *_completed
)
167 bool kvm_make_all_cpus_request(struct kvm
*kvm
, unsigned int req
)
172 struct kvm_vcpu
*vcpu
;
174 zalloc_cpumask_var(&cpus
, GFP_ATOMIC
);
177 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
178 kvm_make_request(req
, vcpu
);
181 /* Set ->requests bit before we read ->mode. */
182 smp_mb__after_atomic();
184 if (cpus
!= NULL
&& cpu
!= -1 && cpu
!= me
&&
185 kvm_vcpu_exiting_guest_mode(vcpu
) != OUTSIDE_GUEST_MODE
)
186 cpumask_set_cpu(cpu
, cpus
);
188 if (unlikely(cpus
== NULL
))
189 smp_call_function_many(cpu_online_mask
, ack_flush
, NULL
, 1);
190 else if (!cpumask_empty(cpus
))
191 smp_call_function_many(cpus
, ack_flush
, NULL
, 1);
195 free_cpumask_var(cpus
);
199 #ifndef CONFIG_HAVE_KVM_ARCH_TLB_FLUSH_ALL
200 void kvm_flush_remote_tlbs(struct kvm
*kvm
)
203 * Read tlbs_dirty before setting KVM_REQ_TLB_FLUSH in
204 * kvm_make_all_cpus_request.
206 long dirty_count
= smp_load_acquire(&kvm
->tlbs_dirty
);
209 * We want to publish modifications to the page tables before reading
210 * mode. Pairs with a memory barrier in arch-specific code.
211 * - x86: smp_mb__after_srcu_read_unlock in vcpu_enter_guest
212 * and smp_mb in walk_shadow_page_lockless_begin/end.
213 * - powerpc: smp_mb in kvmppc_prepare_to_enter.
215 * There is already an smp_mb__after_atomic() before
216 * kvm_make_all_cpus_request() reads vcpu->mode. We reuse that
219 if (kvm_make_all_cpus_request(kvm
, KVM_REQ_TLB_FLUSH
))
220 ++kvm
->stat
.remote_tlb_flush
;
221 cmpxchg(&kvm
->tlbs_dirty
, dirty_count
, 0);
223 EXPORT_SYMBOL_GPL(kvm_flush_remote_tlbs
);
226 void kvm_reload_remote_mmus(struct kvm
*kvm
)
228 kvm_make_all_cpus_request(kvm
, KVM_REQ_MMU_RELOAD
);
231 int kvm_vcpu_init(struct kvm_vcpu
*vcpu
, struct kvm
*kvm
, unsigned id
)
236 mutex_init(&vcpu
->mutex
);
241 init_swait_queue_head(&vcpu
->wq
);
242 kvm_async_pf_vcpu_init(vcpu
);
245 INIT_LIST_HEAD(&vcpu
->blocked_vcpu_list
);
247 page
= alloc_page(GFP_KERNEL
| __GFP_ZERO
);
252 vcpu
->run
= page_address(page
);
254 kvm_vcpu_set_in_spin_loop(vcpu
, false);
255 kvm_vcpu_set_dy_eligible(vcpu
, false);
256 vcpu
->preempted
= false;
258 r
= kvm_arch_vcpu_init(vcpu
);
264 free_page((unsigned long)vcpu
->run
);
268 EXPORT_SYMBOL_GPL(kvm_vcpu_init
);
270 void kvm_vcpu_uninit(struct kvm_vcpu
*vcpu
)
273 kvm_arch_vcpu_uninit(vcpu
);
274 free_page((unsigned long)vcpu
->run
);
276 EXPORT_SYMBOL_GPL(kvm_vcpu_uninit
);
278 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
279 static inline struct kvm
*mmu_notifier_to_kvm(struct mmu_notifier
*mn
)
281 return container_of(mn
, struct kvm
, mmu_notifier
);
284 static void kvm_mmu_notifier_invalidate_page(struct mmu_notifier
*mn
,
285 struct mm_struct
*mm
,
286 unsigned long address
)
288 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
289 int need_tlb_flush
, idx
;
292 * When ->invalidate_page runs, the linux pte has been zapped
293 * already but the page is still allocated until
294 * ->invalidate_page returns. So if we increase the sequence
295 * here the kvm page fault will notice if the spte can't be
296 * established because the page is going to be freed. If
297 * instead the kvm page fault establishes the spte before
298 * ->invalidate_page runs, kvm_unmap_hva will release it
301 * The sequence increase only need to be seen at spin_unlock
302 * time, and not at spin_lock time.
304 * Increasing the sequence after the spin_unlock would be
305 * unsafe because the kvm page fault could then establish the
306 * pte after kvm_unmap_hva returned, without noticing the page
307 * is going to be freed.
309 idx
= srcu_read_lock(&kvm
->srcu
);
310 spin_lock(&kvm
->mmu_lock
);
312 kvm
->mmu_notifier_seq
++;
313 need_tlb_flush
= kvm_unmap_hva(kvm
, address
) | kvm
->tlbs_dirty
;
314 /* we've to flush the tlb before the pages can be freed */
316 kvm_flush_remote_tlbs(kvm
);
318 spin_unlock(&kvm
->mmu_lock
);
320 kvm_arch_mmu_notifier_invalidate_page(kvm
, address
);
322 srcu_read_unlock(&kvm
->srcu
, idx
);
325 static void kvm_mmu_notifier_change_pte(struct mmu_notifier
*mn
,
326 struct mm_struct
*mm
,
327 unsigned long address
,
330 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
333 idx
= srcu_read_lock(&kvm
->srcu
);
334 spin_lock(&kvm
->mmu_lock
);
335 kvm
->mmu_notifier_seq
++;
336 kvm_set_spte_hva(kvm
, address
, pte
);
337 spin_unlock(&kvm
->mmu_lock
);
338 srcu_read_unlock(&kvm
->srcu
, idx
);
341 static void kvm_mmu_notifier_invalidate_range_start(struct mmu_notifier
*mn
,
342 struct mm_struct
*mm
,
346 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
347 int need_tlb_flush
= 0, idx
;
349 idx
= srcu_read_lock(&kvm
->srcu
);
350 spin_lock(&kvm
->mmu_lock
);
352 * The count increase must become visible at unlock time as no
353 * spte can be established without taking the mmu_lock and
354 * count is also read inside the mmu_lock critical section.
356 kvm
->mmu_notifier_count
++;
357 need_tlb_flush
= kvm_unmap_hva_range(kvm
, start
, end
);
358 need_tlb_flush
|= kvm
->tlbs_dirty
;
359 /* we've to flush the tlb before the pages can be freed */
361 kvm_flush_remote_tlbs(kvm
);
363 spin_unlock(&kvm
->mmu_lock
);
364 srcu_read_unlock(&kvm
->srcu
, idx
);
367 static void kvm_mmu_notifier_invalidate_range_end(struct mmu_notifier
*mn
,
368 struct mm_struct
*mm
,
372 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
374 spin_lock(&kvm
->mmu_lock
);
376 * This sequence increase will notify the kvm page fault that
377 * the page that is going to be mapped in the spte could have
380 kvm
->mmu_notifier_seq
++;
383 * The above sequence increase must be visible before the
384 * below count decrease, which is ensured by the smp_wmb above
385 * in conjunction with the smp_rmb in mmu_notifier_retry().
387 kvm
->mmu_notifier_count
--;
388 spin_unlock(&kvm
->mmu_lock
);
390 BUG_ON(kvm
->mmu_notifier_count
< 0);
393 static int kvm_mmu_notifier_clear_flush_young(struct mmu_notifier
*mn
,
394 struct mm_struct
*mm
,
398 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
401 idx
= srcu_read_lock(&kvm
->srcu
);
402 spin_lock(&kvm
->mmu_lock
);
404 young
= kvm_age_hva(kvm
, start
, end
);
406 kvm_flush_remote_tlbs(kvm
);
408 spin_unlock(&kvm
->mmu_lock
);
409 srcu_read_unlock(&kvm
->srcu
, idx
);
414 static int kvm_mmu_notifier_clear_young(struct mmu_notifier
*mn
,
415 struct mm_struct
*mm
,
419 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
422 idx
= srcu_read_lock(&kvm
->srcu
);
423 spin_lock(&kvm
->mmu_lock
);
425 * Even though we do not flush TLB, this will still adversely
426 * affect performance on pre-Haswell Intel EPT, where there is
427 * no EPT Access Bit to clear so that we have to tear down EPT
428 * tables instead. If we find this unacceptable, we can always
429 * add a parameter to kvm_age_hva so that it effectively doesn't
430 * do anything on clear_young.
432 * Also note that currently we never issue secondary TLB flushes
433 * from clear_young, leaving this job up to the regular system
434 * cadence. If we find this inaccurate, we might come up with a
435 * more sophisticated heuristic later.
437 young
= kvm_age_hva(kvm
, start
, end
);
438 spin_unlock(&kvm
->mmu_lock
);
439 srcu_read_unlock(&kvm
->srcu
, idx
);
444 static int kvm_mmu_notifier_test_young(struct mmu_notifier
*mn
,
445 struct mm_struct
*mm
,
446 unsigned long address
)
448 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
451 idx
= srcu_read_lock(&kvm
->srcu
);
452 spin_lock(&kvm
->mmu_lock
);
453 young
= kvm_test_age_hva(kvm
, address
);
454 spin_unlock(&kvm
->mmu_lock
);
455 srcu_read_unlock(&kvm
->srcu
, idx
);
460 static void kvm_mmu_notifier_release(struct mmu_notifier
*mn
,
461 struct mm_struct
*mm
)
463 struct kvm
*kvm
= mmu_notifier_to_kvm(mn
);
466 idx
= srcu_read_lock(&kvm
->srcu
);
467 kvm_arch_flush_shadow_all(kvm
);
468 srcu_read_unlock(&kvm
->srcu
, idx
);
471 static const struct mmu_notifier_ops kvm_mmu_notifier_ops
= {
472 .invalidate_page
= kvm_mmu_notifier_invalidate_page
,
473 .invalidate_range_start
= kvm_mmu_notifier_invalidate_range_start
,
474 .invalidate_range_end
= kvm_mmu_notifier_invalidate_range_end
,
475 .clear_flush_young
= kvm_mmu_notifier_clear_flush_young
,
476 .clear_young
= kvm_mmu_notifier_clear_young
,
477 .test_young
= kvm_mmu_notifier_test_young
,
478 .change_pte
= kvm_mmu_notifier_change_pte
,
479 .release
= kvm_mmu_notifier_release
,
482 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
484 kvm
->mmu_notifier
.ops
= &kvm_mmu_notifier_ops
;
485 return mmu_notifier_register(&kvm
->mmu_notifier
, current
->mm
);
488 #else /* !(CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER) */
490 static int kvm_init_mmu_notifier(struct kvm
*kvm
)
495 #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
497 static struct kvm_memslots
*kvm_alloc_memslots(void)
500 struct kvm_memslots
*slots
;
502 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
507 * Init kvm generation close to the maximum to easily test the
508 * code of handling generation number wrap-around.
510 slots
->generation
= -150;
511 for (i
= 0; i
< KVM_MEM_SLOTS_NUM
; i
++)
512 slots
->id_to_index
[i
] = slots
->memslots
[i
].id
= i
;
517 static void kvm_destroy_dirty_bitmap(struct kvm_memory_slot
*memslot
)
519 if (!memslot
->dirty_bitmap
)
522 kvfree(memslot
->dirty_bitmap
);
523 memslot
->dirty_bitmap
= NULL
;
527 * Free any memory in @free but not in @dont.
529 static void kvm_free_memslot(struct kvm
*kvm
, struct kvm_memory_slot
*free
,
530 struct kvm_memory_slot
*dont
)
532 if (!dont
|| free
->dirty_bitmap
!= dont
->dirty_bitmap
)
533 kvm_destroy_dirty_bitmap(free
);
535 kvm_arch_free_memslot(kvm
, free
, dont
);
540 static void kvm_free_memslots(struct kvm
*kvm
, struct kvm_memslots
*slots
)
542 struct kvm_memory_slot
*memslot
;
547 kvm_for_each_memslot(memslot
, slots
)
548 kvm_free_memslot(kvm
, memslot
, NULL
);
553 static void kvm_destroy_vm_debugfs(struct kvm
*kvm
)
557 if (!kvm
->debugfs_dentry
)
560 debugfs_remove_recursive(kvm
->debugfs_dentry
);
562 for (i
= 0; i
< kvm_debugfs_num_entries
; i
++)
563 kfree(kvm
->debugfs_stat_data
[i
]);
564 kfree(kvm
->debugfs_stat_data
);
567 static int kvm_create_vm_debugfs(struct kvm
*kvm
, int fd
)
569 char dir_name
[ITOA_MAX_LEN
* 2];
570 struct kvm_stat_data
*stat_data
;
571 struct kvm_stats_debugfs_item
*p
;
573 if (!debugfs_initialized())
576 snprintf(dir_name
, sizeof(dir_name
), "%d-%d", task_pid_nr(current
), fd
);
577 kvm
->debugfs_dentry
= debugfs_create_dir(dir_name
,
579 if (!kvm
->debugfs_dentry
)
582 kvm
->debugfs_stat_data
= kcalloc(kvm_debugfs_num_entries
,
583 sizeof(*kvm
->debugfs_stat_data
),
585 if (!kvm
->debugfs_stat_data
)
588 for (p
= debugfs_entries
; p
->name
; p
++) {
589 stat_data
= kzalloc(sizeof(*stat_data
), GFP_KERNEL
);
593 stat_data
->kvm
= kvm
;
594 stat_data
->offset
= p
->offset
;
595 kvm
->debugfs_stat_data
[p
- debugfs_entries
] = stat_data
;
596 if (!debugfs_create_file(p
->name
, 0444,
599 stat_fops_per_vm
[p
->kind
]))
605 static struct kvm
*kvm_create_vm(unsigned long type
)
608 struct kvm
*kvm
= kvm_arch_alloc_vm();
611 return ERR_PTR(-ENOMEM
);
613 spin_lock_init(&kvm
->mmu_lock
);
614 atomic_inc(¤t
->mm
->mm_count
);
615 kvm
->mm
= current
->mm
;
616 kvm_eventfd_init(kvm
);
617 mutex_init(&kvm
->lock
);
618 mutex_init(&kvm
->irq_lock
);
619 mutex_init(&kvm
->slots_lock
);
620 atomic_set(&kvm
->users_count
, 1);
621 INIT_LIST_HEAD(&kvm
->devices
);
623 r
= kvm_arch_init_vm(kvm
, type
);
625 goto out_err_no_disable
;
627 r
= hardware_enable_all();
629 goto out_err_no_disable
;
631 #ifdef CONFIG_HAVE_KVM_IRQFD
632 INIT_HLIST_HEAD(&kvm
->irq_ack_notifier_list
);
635 BUILD_BUG_ON(KVM_MEM_SLOTS_NUM
> SHRT_MAX
);
638 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++) {
639 kvm
->memslots
[i
] = kvm_alloc_memslots();
640 if (!kvm
->memslots
[i
])
641 goto out_err_no_srcu
;
644 if (init_srcu_struct(&kvm
->srcu
))
645 goto out_err_no_srcu
;
646 if (init_srcu_struct(&kvm
->irq_srcu
))
647 goto out_err_no_irq_srcu
;
648 for (i
= 0; i
< KVM_NR_BUSES
; i
++) {
649 kvm
->buses
[i
] = kzalloc(sizeof(struct kvm_io_bus
),
655 r
= kvm_init_mmu_notifier(kvm
);
659 spin_lock(&kvm_lock
);
660 list_add(&kvm
->vm_list
, &vm_list
);
661 spin_unlock(&kvm_lock
);
663 preempt_notifier_inc();
668 cleanup_srcu_struct(&kvm
->irq_srcu
);
670 cleanup_srcu_struct(&kvm
->srcu
);
672 hardware_disable_all();
674 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
675 kfree(kvm
->buses
[i
]);
676 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
677 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
678 kvm_arch_free_vm(kvm
);
684 * Avoid using vmalloc for a small buffer.
685 * Should not be used when the size is statically known.
687 void *kvm_kvzalloc(unsigned long size
)
689 if (size
> PAGE_SIZE
)
690 return vzalloc(size
);
692 return kzalloc(size
, GFP_KERNEL
);
695 static void kvm_destroy_devices(struct kvm
*kvm
)
697 struct kvm_device
*dev
, *tmp
;
699 list_for_each_entry_safe(dev
, tmp
, &kvm
->devices
, vm_node
) {
700 list_del(&dev
->vm_node
);
701 dev
->ops
->destroy(dev
);
705 static void kvm_destroy_vm(struct kvm
*kvm
)
708 struct mm_struct
*mm
= kvm
->mm
;
710 kvm_destroy_vm_debugfs(kvm
);
711 kvm_arch_sync_events(kvm
);
712 spin_lock(&kvm_lock
);
713 list_del(&kvm
->vm_list
);
714 spin_unlock(&kvm_lock
);
715 kvm_free_irq_routing(kvm
);
716 for (i
= 0; i
< KVM_NR_BUSES
; i
++)
717 kvm_io_bus_destroy(kvm
->buses
[i
]);
718 kvm_coalesced_mmio_free(kvm
);
719 #if defined(CONFIG_MMU_NOTIFIER) && defined(KVM_ARCH_WANT_MMU_NOTIFIER)
720 mmu_notifier_unregister(&kvm
->mmu_notifier
, kvm
->mm
);
722 kvm_arch_flush_shadow_all(kvm
);
724 kvm_arch_destroy_vm(kvm
);
725 kvm_destroy_devices(kvm
);
726 for (i
= 0; i
< KVM_ADDRESS_SPACE_NUM
; i
++)
727 kvm_free_memslots(kvm
, kvm
->memslots
[i
]);
728 cleanup_srcu_struct(&kvm
->irq_srcu
);
729 cleanup_srcu_struct(&kvm
->srcu
);
730 kvm_arch_free_vm(kvm
);
731 preempt_notifier_dec();
732 hardware_disable_all();
736 void kvm_get_kvm(struct kvm
*kvm
)
738 atomic_inc(&kvm
->users_count
);
740 EXPORT_SYMBOL_GPL(kvm_get_kvm
);
742 void kvm_put_kvm(struct kvm
*kvm
)
744 if (atomic_dec_and_test(&kvm
->users_count
))
747 EXPORT_SYMBOL_GPL(kvm_put_kvm
);
750 static int kvm_vm_release(struct inode
*inode
, struct file
*filp
)
752 struct kvm
*kvm
= filp
->private_data
;
754 kvm_irqfd_release(kvm
);
761 * Allocation size is twice as large as the actual dirty bitmap size.
762 * See x86's kvm_vm_ioctl_get_dirty_log() why this is needed.
764 static int kvm_create_dirty_bitmap(struct kvm_memory_slot
*memslot
)
766 unsigned long dirty_bytes
= 2 * kvm_dirty_bitmap_bytes(memslot
);
768 memslot
->dirty_bitmap
= kvm_kvzalloc(dirty_bytes
);
769 if (!memslot
->dirty_bitmap
)
776 * Insert memslot and re-sort memslots based on their GFN,
777 * so binary search could be used to lookup GFN.
778 * Sorting algorithm takes advantage of having initially
779 * sorted array and known changed memslot position.
781 static void update_memslots(struct kvm_memslots
*slots
,
782 struct kvm_memory_slot
*new)
785 int i
= slots
->id_to_index
[id
];
786 struct kvm_memory_slot
*mslots
= slots
->memslots
;
788 WARN_ON(mslots
[i
].id
!= id
);
790 WARN_ON(!mslots
[i
].npages
);
791 if (mslots
[i
].npages
)
794 if (!mslots
[i
].npages
)
798 while (i
< KVM_MEM_SLOTS_NUM
- 1 &&
799 new->base_gfn
<= mslots
[i
+ 1].base_gfn
) {
800 if (!mslots
[i
+ 1].npages
)
802 mslots
[i
] = mslots
[i
+ 1];
803 slots
->id_to_index
[mslots
[i
].id
] = i
;
808 * The ">=" is needed when creating a slot with base_gfn == 0,
809 * so that it moves before all those with base_gfn == npages == 0.
811 * On the other hand, if new->npages is zero, the above loop has
812 * already left i pointing to the beginning of the empty part of
813 * mslots, and the ">=" would move the hole backwards in this
814 * case---which is wrong. So skip the loop when deleting a slot.
818 new->base_gfn
>= mslots
[i
- 1].base_gfn
) {
819 mslots
[i
] = mslots
[i
- 1];
820 slots
->id_to_index
[mslots
[i
].id
] = i
;
824 WARN_ON_ONCE(i
!= slots
->used_slots
);
827 slots
->id_to_index
[mslots
[i
].id
] = i
;
830 static int check_memory_region_flags(const struct kvm_userspace_memory_region
*mem
)
832 u32 valid_flags
= KVM_MEM_LOG_DIRTY_PAGES
;
834 #ifdef __KVM_HAVE_READONLY_MEM
835 valid_flags
|= KVM_MEM_READONLY
;
838 if (mem
->flags
& ~valid_flags
)
844 static struct kvm_memslots
*install_new_memslots(struct kvm
*kvm
,
845 int as_id
, struct kvm_memslots
*slots
)
847 struct kvm_memslots
*old_memslots
= __kvm_memslots(kvm
, as_id
);
850 * Set the low bit in the generation, which disables SPTE caching
851 * until the end of synchronize_srcu_expedited.
853 WARN_ON(old_memslots
->generation
& 1);
854 slots
->generation
= old_memslots
->generation
+ 1;
856 rcu_assign_pointer(kvm
->memslots
[as_id
], slots
);
857 synchronize_srcu_expedited(&kvm
->srcu
);
860 * Increment the new memslot generation a second time. This prevents
861 * vm exits that race with memslot updates from caching a memslot
862 * generation that will (potentially) be valid forever.
866 kvm_arch_memslots_updated(kvm
, slots
);
872 * Allocate some memory and give it an address in the guest physical address
875 * Discontiguous memory is allowed, mostly for framebuffers.
877 * Must be called holding kvm->slots_lock for write.
879 int __kvm_set_memory_region(struct kvm
*kvm
,
880 const struct kvm_userspace_memory_region
*mem
)
884 unsigned long npages
;
885 struct kvm_memory_slot
*slot
;
886 struct kvm_memory_slot old
, new;
887 struct kvm_memslots
*slots
= NULL
, *old_memslots
;
889 enum kvm_mr_change change
;
891 r
= check_memory_region_flags(mem
);
896 as_id
= mem
->slot
>> 16;
899 /* General sanity checks */
900 if (mem
->memory_size
& (PAGE_SIZE
- 1))
902 if (mem
->guest_phys_addr
& (PAGE_SIZE
- 1))
904 /* We can read the guest memory with __xxx_user() later on. */
905 if ((id
< KVM_USER_MEM_SLOTS
) &&
906 ((mem
->userspace_addr
& (PAGE_SIZE
- 1)) ||
907 !access_ok(VERIFY_WRITE
,
908 (void __user
*)(unsigned long)mem
->userspace_addr
,
911 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_MEM_SLOTS_NUM
)
913 if (mem
->guest_phys_addr
+ mem
->memory_size
< mem
->guest_phys_addr
)
916 slot
= id_to_memslot(__kvm_memslots(kvm
, as_id
), id
);
917 base_gfn
= mem
->guest_phys_addr
>> PAGE_SHIFT
;
918 npages
= mem
->memory_size
>> PAGE_SHIFT
;
920 if (npages
> KVM_MEM_MAX_NR_PAGES
)
926 new.base_gfn
= base_gfn
;
928 new.flags
= mem
->flags
;
932 change
= KVM_MR_CREATE
;
933 else { /* Modify an existing slot. */
934 if ((mem
->userspace_addr
!= old
.userspace_addr
) ||
935 (npages
!= old
.npages
) ||
936 ((new.flags
^ old
.flags
) & KVM_MEM_READONLY
))
939 if (base_gfn
!= old
.base_gfn
)
940 change
= KVM_MR_MOVE
;
941 else if (new.flags
!= old
.flags
)
942 change
= KVM_MR_FLAGS_ONLY
;
943 else { /* Nothing to change. */
952 change
= KVM_MR_DELETE
;
957 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
958 /* Check for overlaps */
960 kvm_for_each_memslot(slot
, __kvm_memslots(kvm
, as_id
)) {
961 if ((slot
->id
>= KVM_USER_MEM_SLOTS
) ||
964 if (!((base_gfn
+ npages
<= slot
->base_gfn
) ||
965 (base_gfn
>= slot
->base_gfn
+ slot
->npages
)))
970 /* Free page dirty bitmap if unneeded */
971 if (!(new.flags
& KVM_MEM_LOG_DIRTY_PAGES
))
972 new.dirty_bitmap
= NULL
;
975 if (change
== KVM_MR_CREATE
) {
976 new.userspace_addr
= mem
->userspace_addr
;
978 if (kvm_arch_create_memslot(kvm
, &new, npages
))
982 /* Allocate page dirty bitmap if needed */
983 if ((new.flags
& KVM_MEM_LOG_DIRTY_PAGES
) && !new.dirty_bitmap
) {
984 if (kvm_create_dirty_bitmap(&new) < 0)
988 slots
= kvm_kvzalloc(sizeof(struct kvm_memslots
));
991 memcpy(slots
, __kvm_memslots(kvm
, as_id
), sizeof(struct kvm_memslots
));
993 if ((change
== KVM_MR_DELETE
) || (change
== KVM_MR_MOVE
)) {
994 slot
= id_to_memslot(slots
, id
);
995 slot
->flags
|= KVM_MEMSLOT_INVALID
;
997 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
999 /* slot was deleted or moved, clear iommu mapping */
1000 kvm_iommu_unmap_pages(kvm
, &old
);
1001 /* From this point no new shadow pages pointing to a deleted,
1002 * or moved, memslot will be created.
1004 * validation of sp->gfn happens in:
1005 * - gfn_to_hva (kvm_read_guest, gfn_to_pfn)
1006 * - kvm_is_visible_gfn (mmu_check_roots)
1008 kvm_arch_flush_shadow_memslot(kvm
, slot
);
1011 * We can re-use the old_memslots from above, the only difference
1012 * from the currently installed memslots is the invalid flag. This
1013 * will get overwritten by update_memslots anyway.
1015 slots
= old_memslots
;
1018 r
= kvm_arch_prepare_memory_region(kvm
, &new, mem
, change
);
1022 /* actual memory is freed via old in kvm_free_memslot below */
1023 if (change
== KVM_MR_DELETE
) {
1024 new.dirty_bitmap
= NULL
;
1025 memset(&new.arch
, 0, sizeof(new.arch
));
1028 update_memslots(slots
, &new);
1029 old_memslots
= install_new_memslots(kvm
, as_id
, slots
);
1031 kvm_arch_commit_memory_region(kvm
, mem
, &old
, &new, change
);
1033 kvm_free_memslot(kvm
, &old
, &new);
1034 kvfree(old_memslots
);
1037 * IOMMU mapping: New slots need to be mapped. Old slots need to be
1038 * un-mapped and re-mapped if their base changes. Since base change
1039 * unmapping is handled above with slot deletion, mapping alone is
1040 * needed here. Anything else the iommu might care about for existing
1041 * slots (size changes, userspace addr changes and read-only flag
1042 * changes) is disallowed above, so any other attribute changes getting
1043 * here can be skipped.
1045 if ((change
== KVM_MR_CREATE
) || (change
== KVM_MR_MOVE
)) {
1046 r
= kvm_iommu_map_pages(kvm
, &new);
1055 kvm_free_memslot(kvm
, &new, &old
);
1059 EXPORT_SYMBOL_GPL(__kvm_set_memory_region
);
1061 int kvm_set_memory_region(struct kvm
*kvm
,
1062 const struct kvm_userspace_memory_region
*mem
)
1066 mutex_lock(&kvm
->slots_lock
);
1067 r
= __kvm_set_memory_region(kvm
, mem
);
1068 mutex_unlock(&kvm
->slots_lock
);
1071 EXPORT_SYMBOL_GPL(kvm_set_memory_region
);
1073 static int kvm_vm_ioctl_set_memory_region(struct kvm
*kvm
,
1074 struct kvm_userspace_memory_region
*mem
)
1076 if ((u16
)mem
->slot
>= KVM_USER_MEM_SLOTS
)
1079 return kvm_set_memory_region(kvm
, mem
);
1082 int kvm_get_dirty_log(struct kvm
*kvm
,
1083 struct kvm_dirty_log
*log
, int *is_dirty
)
1085 struct kvm_memslots
*slots
;
1086 struct kvm_memory_slot
*memslot
;
1087 int r
, i
, as_id
, id
;
1089 unsigned long any
= 0;
1092 as_id
= log
->slot
>> 16;
1093 id
= (u16
)log
->slot
;
1094 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1097 slots
= __kvm_memslots(kvm
, as_id
);
1098 memslot
= id_to_memslot(slots
, id
);
1100 if (!memslot
->dirty_bitmap
)
1103 n
= kvm_dirty_bitmap_bytes(memslot
);
1105 for (i
= 0; !any
&& i
< n
/sizeof(long); ++i
)
1106 any
= memslot
->dirty_bitmap
[i
];
1109 if (copy_to_user(log
->dirty_bitmap
, memslot
->dirty_bitmap
, n
))
1119 EXPORT_SYMBOL_GPL(kvm_get_dirty_log
);
1121 #ifdef CONFIG_KVM_GENERIC_DIRTYLOG_READ_PROTECT
1123 * kvm_get_dirty_log_protect - get a snapshot of dirty pages, and if any pages
1124 * are dirty write protect them for next write.
1125 * @kvm: pointer to kvm instance
1126 * @log: slot id and address to which we copy the log
1127 * @is_dirty: flag set if any page is dirty
1129 * We need to keep it in mind that VCPU threads can write to the bitmap
1130 * concurrently. So, to avoid losing track of dirty pages we keep the
1133 * 1. Take a snapshot of the bit and clear it if needed.
1134 * 2. Write protect the corresponding page.
1135 * 3. Copy the snapshot to the userspace.
1136 * 4. Upon return caller flushes TLB's if needed.
1138 * Between 2 and 4, the guest may write to the page using the remaining TLB
1139 * entry. This is not a problem because the page is reported dirty using
1140 * the snapshot taken before and step 4 ensures that writes done after
1141 * exiting to userspace will be logged for the next call.
1144 int kvm_get_dirty_log_protect(struct kvm
*kvm
,
1145 struct kvm_dirty_log
*log
, bool *is_dirty
)
1147 struct kvm_memslots
*slots
;
1148 struct kvm_memory_slot
*memslot
;
1149 int r
, i
, as_id
, id
;
1151 unsigned long *dirty_bitmap
;
1152 unsigned long *dirty_bitmap_buffer
;
1155 as_id
= log
->slot
>> 16;
1156 id
= (u16
)log
->slot
;
1157 if (as_id
>= KVM_ADDRESS_SPACE_NUM
|| id
>= KVM_USER_MEM_SLOTS
)
1160 slots
= __kvm_memslots(kvm
, as_id
);
1161 memslot
= id_to_memslot(slots
, id
);
1163 dirty_bitmap
= memslot
->dirty_bitmap
;
1168 n
= kvm_dirty_bitmap_bytes(memslot
);
1170 dirty_bitmap_buffer
= dirty_bitmap
+ n
/ sizeof(long);
1171 memset(dirty_bitmap_buffer
, 0, n
);
1173 spin_lock(&kvm
->mmu_lock
);
1175 for (i
= 0; i
< n
/ sizeof(long); i
++) {
1179 if (!dirty_bitmap
[i
])
1184 mask
= xchg(&dirty_bitmap
[i
], 0);
1185 dirty_bitmap_buffer
[i
] = mask
;
1188 offset
= i
* BITS_PER_LONG
;
1189 kvm_arch_mmu_enable_log_dirty_pt_masked(kvm
, memslot
,
1194 spin_unlock(&kvm
->mmu_lock
);
1197 if (copy_to_user(log
->dirty_bitmap
, dirty_bitmap_buffer
, n
))
1204 EXPORT_SYMBOL_GPL(kvm_get_dirty_log_protect
);
1207 bool kvm_largepages_enabled(void)
1209 return largepages_enabled
;
1212 void kvm_disable_largepages(void)
1214 largepages_enabled
= false;
1216 EXPORT_SYMBOL_GPL(kvm_disable_largepages
);
1218 struct kvm_memory_slot
*gfn_to_memslot(struct kvm
*kvm
, gfn_t gfn
)
1220 return __gfn_to_memslot(kvm_memslots(kvm
), gfn
);
1222 EXPORT_SYMBOL_GPL(gfn_to_memslot
);
1224 struct kvm_memory_slot
*kvm_vcpu_gfn_to_memslot(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1226 return __gfn_to_memslot(kvm_vcpu_memslots(vcpu
), gfn
);
1229 bool kvm_is_visible_gfn(struct kvm
*kvm
, gfn_t gfn
)
1231 struct kvm_memory_slot
*memslot
= gfn_to_memslot(kvm
, gfn
);
1233 if (!memslot
|| memslot
->id
>= KVM_USER_MEM_SLOTS
||
1234 memslot
->flags
& KVM_MEMSLOT_INVALID
)
1239 EXPORT_SYMBOL_GPL(kvm_is_visible_gfn
);
1241 unsigned long kvm_host_page_size(struct kvm
*kvm
, gfn_t gfn
)
1243 struct vm_area_struct
*vma
;
1244 unsigned long addr
, size
;
1248 addr
= gfn_to_hva(kvm
, gfn
);
1249 if (kvm_is_error_hva(addr
))
1252 down_read(¤t
->mm
->mmap_sem
);
1253 vma
= find_vma(current
->mm
, addr
);
1257 size
= vma_kernel_pagesize(vma
);
1260 up_read(¤t
->mm
->mmap_sem
);
1265 static bool memslot_is_readonly(struct kvm_memory_slot
*slot
)
1267 return slot
->flags
& KVM_MEM_READONLY
;
1270 static unsigned long __gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1271 gfn_t
*nr_pages
, bool write
)
1273 if (!slot
|| slot
->flags
& KVM_MEMSLOT_INVALID
)
1274 return KVM_HVA_ERR_BAD
;
1276 if (memslot_is_readonly(slot
) && write
)
1277 return KVM_HVA_ERR_RO_BAD
;
1280 *nr_pages
= slot
->npages
- (gfn
- slot
->base_gfn
);
1282 return __gfn_to_hva_memslot(slot
, gfn
);
1285 static unsigned long gfn_to_hva_many(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1288 return __gfn_to_hva_many(slot
, gfn
, nr_pages
, true);
1291 unsigned long gfn_to_hva_memslot(struct kvm_memory_slot
*slot
,
1294 return gfn_to_hva_many(slot
, gfn
, NULL
);
1296 EXPORT_SYMBOL_GPL(gfn_to_hva_memslot
);
1298 unsigned long gfn_to_hva(struct kvm
*kvm
, gfn_t gfn
)
1300 return gfn_to_hva_many(gfn_to_memslot(kvm
, gfn
), gfn
, NULL
);
1302 EXPORT_SYMBOL_GPL(gfn_to_hva
);
1304 unsigned long kvm_vcpu_gfn_to_hva(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1306 return gfn_to_hva_many(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
, NULL
);
1308 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_hva
);
1311 * If writable is set to false, the hva returned by this function is only
1312 * allowed to be read.
1314 unsigned long gfn_to_hva_memslot_prot(struct kvm_memory_slot
*slot
,
1315 gfn_t gfn
, bool *writable
)
1317 unsigned long hva
= __gfn_to_hva_many(slot
, gfn
, NULL
, false);
1319 if (!kvm_is_error_hva(hva
) && writable
)
1320 *writable
= !memslot_is_readonly(slot
);
1325 unsigned long gfn_to_hva_prot(struct kvm
*kvm
, gfn_t gfn
, bool *writable
)
1327 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1329 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1332 unsigned long kvm_vcpu_gfn_to_hva_prot(struct kvm_vcpu
*vcpu
, gfn_t gfn
, bool *writable
)
1334 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1336 return gfn_to_hva_memslot_prot(slot
, gfn
, writable
);
1339 static int get_user_page_nowait(unsigned long start
, int write
,
1342 int flags
= FOLL_TOUCH
| FOLL_NOWAIT
| FOLL_HWPOISON
| FOLL_GET
;
1345 flags
|= FOLL_WRITE
;
1347 return __get_user_pages(current
, current
->mm
, start
, 1, flags
, page
,
1351 static inline int check_user_page_hwpoison(unsigned long addr
)
1353 int rc
, flags
= FOLL_TOUCH
| FOLL_HWPOISON
| FOLL_WRITE
;
1355 rc
= __get_user_pages(current
, current
->mm
, addr
, 1,
1356 flags
, NULL
, NULL
, NULL
);
1357 return rc
== -EHWPOISON
;
1361 * The atomic path to get the writable pfn which will be stored in @pfn,
1362 * true indicates success, otherwise false is returned.
1364 static bool hva_to_pfn_fast(unsigned long addr
, bool atomic
, bool *async
,
1365 bool write_fault
, bool *writable
, kvm_pfn_t
*pfn
)
1367 struct page
*page
[1];
1370 if (!(async
|| atomic
))
1374 * Fast pin a writable pfn only if it is a write fault request
1375 * or the caller allows to map a writable pfn for a read fault
1378 if (!(write_fault
|| writable
))
1381 npages
= __get_user_pages_fast(addr
, 1, 1, page
);
1383 *pfn
= page_to_pfn(page
[0]);
1394 * The slow path to get the pfn of the specified host virtual address,
1395 * 1 indicates success, -errno is returned if error is detected.
1397 static int hva_to_pfn_slow(unsigned long addr
, bool *async
, bool write_fault
,
1398 bool *writable
, kvm_pfn_t
*pfn
)
1400 struct page
*page
[1];
1406 *writable
= write_fault
;
1409 down_read(¤t
->mm
->mmap_sem
);
1410 npages
= get_user_page_nowait(addr
, write_fault
, page
);
1411 up_read(¤t
->mm
->mmap_sem
);
1413 npages
= __get_user_pages_unlocked(current
, current
->mm
, addr
, 1,
1414 write_fault
, 0, page
,
1415 FOLL_TOUCH
|FOLL_HWPOISON
);
1419 /* map read fault as writable if possible */
1420 if (unlikely(!write_fault
) && writable
) {
1421 struct page
*wpage
[1];
1423 npages
= __get_user_pages_fast(addr
, 1, 1, wpage
);
1432 *pfn
= page_to_pfn(page
[0]);
1436 static bool vma_is_valid(struct vm_area_struct
*vma
, bool write_fault
)
1438 if (unlikely(!(vma
->vm_flags
& VM_READ
)))
1441 if (write_fault
&& (unlikely(!(vma
->vm_flags
& VM_WRITE
))))
1447 static int hva_to_pfn_remapped(struct vm_area_struct
*vma
,
1448 unsigned long addr
, bool *async
,
1449 bool write_fault
, kvm_pfn_t
*p_pfn
)
1454 r
= follow_pfn(vma
, addr
, &pfn
);
1457 * get_user_pages fails for VM_IO and VM_PFNMAP vmas and does
1458 * not call the fault handler, so do it here.
1460 bool unlocked
= false;
1461 r
= fixup_user_fault(current
, current
->mm
, addr
,
1462 (write_fault
? FAULT_FLAG_WRITE
: 0),
1469 r
= follow_pfn(vma
, addr
, &pfn
);
1477 * Get a reference here because callers of *hva_to_pfn* and
1478 * *gfn_to_pfn* ultimately call kvm_release_pfn_clean on the
1479 * returned pfn. This is only needed if the VMA has VM_MIXEDMAP
1480 * set, but the kvm_get_pfn/kvm_release_pfn_clean pair will
1481 * simply do nothing for reserved pfns.
1483 * Whoever called remap_pfn_range is also going to call e.g.
1484 * unmap_mapping_range before the underlying pages are freed,
1485 * causing a call to our MMU notifier.
1494 * Pin guest page in memory and return its pfn.
1495 * @addr: host virtual address which maps memory to the guest
1496 * @atomic: whether this function can sleep
1497 * @async: whether this function need to wait IO complete if the
1498 * host page is not in the memory
1499 * @write_fault: whether we should get a writable host page
1500 * @writable: whether it allows to map a writable host page for !@write_fault
1502 * The function will map a writable host page for these two cases:
1503 * 1): @write_fault = true
1504 * 2): @write_fault = false && @writable, @writable will tell the caller
1505 * whether the mapping is writable.
1507 static kvm_pfn_t
hva_to_pfn(unsigned long addr
, bool atomic
, bool *async
,
1508 bool write_fault
, bool *writable
)
1510 struct vm_area_struct
*vma
;
1514 /* we can do it either atomically or asynchronously, not both */
1515 BUG_ON(atomic
&& async
);
1517 if (hva_to_pfn_fast(addr
, atomic
, async
, write_fault
, writable
, &pfn
))
1521 return KVM_PFN_ERR_FAULT
;
1523 npages
= hva_to_pfn_slow(addr
, async
, write_fault
, writable
, &pfn
);
1527 down_read(¤t
->mm
->mmap_sem
);
1528 if (npages
== -EHWPOISON
||
1529 (!async
&& check_user_page_hwpoison(addr
))) {
1530 pfn
= KVM_PFN_ERR_HWPOISON
;
1535 vma
= find_vma_intersection(current
->mm
, addr
, addr
+ 1);
1538 pfn
= KVM_PFN_ERR_FAULT
;
1539 else if (vma
->vm_flags
& (VM_IO
| VM_PFNMAP
)) {
1540 r
= hva_to_pfn_remapped(vma
, addr
, async
, write_fault
, &pfn
);
1544 pfn
= KVM_PFN_ERR_FAULT
;
1546 if (async
&& vma_is_valid(vma
, write_fault
))
1548 pfn
= KVM_PFN_ERR_FAULT
;
1551 up_read(¤t
->mm
->mmap_sem
);
1555 kvm_pfn_t
__gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1556 bool atomic
, bool *async
, bool write_fault
,
1559 unsigned long addr
= __gfn_to_hva_many(slot
, gfn
, NULL
, write_fault
);
1561 if (addr
== KVM_HVA_ERR_RO_BAD
) {
1564 return KVM_PFN_ERR_RO_FAULT
;
1567 if (kvm_is_error_hva(addr
)) {
1570 return KVM_PFN_NOSLOT
;
1573 /* Do not map writable pfn in the readonly memslot. */
1574 if (writable
&& memslot_is_readonly(slot
)) {
1579 return hva_to_pfn(addr
, atomic
, async
, write_fault
,
1582 EXPORT_SYMBOL_GPL(__gfn_to_pfn_memslot
);
1584 kvm_pfn_t
gfn_to_pfn_prot(struct kvm
*kvm
, gfn_t gfn
, bool write_fault
,
1587 return __gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
, false, NULL
,
1588 write_fault
, writable
);
1590 EXPORT_SYMBOL_GPL(gfn_to_pfn_prot
);
1592 kvm_pfn_t
gfn_to_pfn_memslot(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1594 return __gfn_to_pfn_memslot(slot
, gfn
, false, NULL
, true, NULL
);
1596 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot
);
1598 kvm_pfn_t
gfn_to_pfn_memslot_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
)
1600 return __gfn_to_pfn_memslot(slot
, gfn
, true, NULL
, true, NULL
);
1602 EXPORT_SYMBOL_GPL(gfn_to_pfn_memslot_atomic
);
1604 kvm_pfn_t
gfn_to_pfn_atomic(struct kvm
*kvm
, gfn_t gfn
)
1606 return gfn_to_pfn_memslot_atomic(gfn_to_memslot(kvm
, gfn
), gfn
);
1608 EXPORT_SYMBOL_GPL(gfn_to_pfn_atomic
);
1610 kvm_pfn_t
kvm_vcpu_gfn_to_pfn_atomic(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1612 return gfn_to_pfn_memslot_atomic(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1614 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn_atomic
);
1616 kvm_pfn_t
gfn_to_pfn(struct kvm
*kvm
, gfn_t gfn
)
1618 return gfn_to_pfn_memslot(gfn_to_memslot(kvm
, gfn
), gfn
);
1620 EXPORT_SYMBOL_GPL(gfn_to_pfn
);
1622 kvm_pfn_t
kvm_vcpu_gfn_to_pfn(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1624 return gfn_to_pfn_memslot(kvm_vcpu_gfn_to_memslot(vcpu
, gfn
), gfn
);
1626 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_pfn
);
1628 int gfn_to_page_many_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1629 struct page
**pages
, int nr_pages
)
1634 addr
= gfn_to_hva_many(slot
, gfn
, &entry
);
1635 if (kvm_is_error_hva(addr
))
1638 if (entry
< nr_pages
)
1641 return __get_user_pages_fast(addr
, nr_pages
, 1, pages
);
1643 EXPORT_SYMBOL_GPL(gfn_to_page_many_atomic
);
1645 static struct page
*kvm_pfn_to_page(kvm_pfn_t pfn
)
1647 if (is_error_noslot_pfn(pfn
))
1648 return KVM_ERR_PTR_BAD_PAGE
;
1650 if (kvm_is_reserved_pfn(pfn
)) {
1652 return KVM_ERR_PTR_BAD_PAGE
;
1655 return pfn_to_page(pfn
);
1658 struct page
*gfn_to_page(struct kvm
*kvm
, gfn_t gfn
)
1662 pfn
= gfn_to_pfn(kvm
, gfn
);
1664 return kvm_pfn_to_page(pfn
);
1666 EXPORT_SYMBOL_GPL(gfn_to_page
);
1668 struct page
*kvm_vcpu_gfn_to_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
1672 pfn
= kvm_vcpu_gfn_to_pfn(vcpu
, gfn
);
1674 return kvm_pfn_to_page(pfn
);
1676 EXPORT_SYMBOL_GPL(kvm_vcpu_gfn_to_page
);
1678 void kvm_release_page_clean(struct page
*page
)
1680 WARN_ON(is_error_page(page
));
1682 kvm_release_pfn_clean(page_to_pfn(page
));
1684 EXPORT_SYMBOL_GPL(kvm_release_page_clean
);
1686 void kvm_release_pfn_clean(kvm_pfn_t pfn
)
1688 if (!is_error_noslot_pfn(pfn
) && !kvm_is_reserved_pfn(pfn
))
1689 put_page(pfn_to_page(pfn
));
1691 EXPORT_SYMBOL_GPL(kvm_release_pfn_clean
);
1693 void kvm_release_page_dirty(struct page
*page
)
1695 WARN_ON(is_error_page(page
));
1697 kvm_release_pfn_dirty(page_to_pfn(page
));
1699 EXPORT_SYMBOL_GPL(kvm_release_page_dirty
);
1701 static void kvm_release_pfn_dirty(kvm_pfn_t pfn
)
1703 kvm_set_pfn_dirty(pfn
);
1704 kvm_release_pfn_clean(pfn
);
1707 void kvm_set_pfn_dirty(kvm_pfn_t pfn
)
1709 if (!kvm_is_reserved_pfn(pfn
)) {
1710 struct page
*page
= pfn_to_page(pfn
);
1712 if (!PageReserved(page
))
1716 EXPORT_SYMBOL_GPL(kvm_set_pfn_dirty
);
1718 void kvm_set_pfn_accessed(kvm_pfn_t pfn
)
1720 if (!kvm_is_reserved_pfn(pfn
))
1721 mark_page_accessed(pfn_to_page(pfn
));
1723 EXPORT_SYMBOL_GPL(kvm_set_pfn_accessed
);
1725 void kvm_get_pfn(kvm_pfn_t pfn
)
1727 if (!kvm_is_reserved_pfn(pfn
))
1728 get_page(pfn_to_page(pfn
));
1730 EXPORT_SYMBOL_GPL(kvm_get_pfn
);
1732 static int next_segment(unsigned long len
, int offset
)
1734 if (len
> PAGE_SIZE
- offset
)
1735 return PAGE_SIZE
- offset
;
1740 static int __kvm_read_guest_page(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1741 void *data
, int offset
, int len
)
1746 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1747 if (kvm_is_error_hva(addr
))
1749 r
= __copy_from_user(data
, (void __user
*)addr
+ offset
, len
);
1755 int kvm_read_guest_page(struct kvm
*kvm
, gfn_t gfn
, void *data
, int offset
,
1758 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1760 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1762 EXPORT_SYMBOL_GPL(kvm_read_guest_page
);
1764 int kvm_vcpu_read_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
, void *data
,
1765 int offset
, int len
)
1767 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1769 return __kvm_read_guest_page(slot
, gfn
, data
, offset
, len
);
1771 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_page
);
1773 int kvm_read_guest(struct kvm
*kvm
, gpa_t gpa
, void *data
, unsigned long len
)
1775 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1777 int offset
= offset_in_page(gpa
);
1780 while ((seg
= next_segment(len
, offset
)) != 0) {
1781 ret
= kvm_read_guest_page(kvm
, gfn
, data
, offset
, seg
);
1791 EXPORT_SYMBOL_GPL(kvm_read_guest
);
1793 int kvm_vcpu_read_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, void *data
, unsigned long len
)
1795 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1797 int offset
= offset_in_page(gpa
);
1800 while ((seg
= next_segment(len
, offset
)) != 0) {
1801 ret
= kvm_vcpu_read_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1811 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest
);
1813 static int __kvm_read_guest_atomic(struct kvm_memory_slot
*slot
, gfn_t gfn
,
1814 void *data
, int offset
, unsigned long len
)
1819 addr
= gfn_to_hva_memslot_prot(slot
, gfn
, NULL
);
1820 if (kvm_is_error_hva(addr
))
1822 pagefault_disable();
1823 r
= __copy_from_user_inatomic(data
, (void __user
*)addr
+ offset
, len
);
1830 int kvm_read_guest_atomic(struct kvm
*kvm
, gpa_t gpa
, void *data
,
1833 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1834 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1835 int offset
= offset_in_page(gpa
);
1837 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1839 EXPORT_SYMBOL_GPL(kvm_read_guest_atomic
);
1841 int kvm_vcpu_read_guest_atomic(struct kvm_vcpu
*vcpu
, gpa_t gpa
,
1842 void *data
, unsigned long len
)
1844 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1845 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1846 int offset
= offset_in_page(gpa
);
1848 return __kvm_read_guest_atomic(slot
, gfn
, data
, offset
, len
);
1850 EXPORT_SYMBOL_GPL(kvm_vcpu_read_guest_atomic
);
1852 static int __kvm_write_guest_page(struct kvm_memory_slot
*memslot
, gfn_t gfn
,
1853 const void *data
, int offset
, int len
)
1858 addr
= gfn_to_hva_memslot(memslot
, gfn
);
1859 if (kvm_is_error_hva(addr
))
1861 r
= __copy_to_user((void __user
*)addr
+ offset
, data
, len
);
1864 mark_page_dirty_in_slot(memslot
, gfn
);
1868 int kvm_write_guest_page(struct kvm
*kvm
, gfn_t gfn
,
1869 const void *data
, int offset
, int len
)
1871 struct kvm_memory_slot
*slot
= gfn_to_memslot(kvm
, gfn
);
1873 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1875 EXPORT_SYMBOL_GPL(kvm_write_guest_page
);
1877 int kvm_vcpu_write_guest_page(struct kvm_vcpu
*vcpu
, gfn_t gfn
,
1878 const void *data
, int offset
, int len
)
1880 struct kvm_memory_slot
*slot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
1882 return __kvm_write_guest_page(slot
, gfn
, data
, offset
, len
);
1884 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest_page
);
1886 int kvm_write_guest(struct kvm
*kvm
, gpa_t gpa
, const void *data
,
1889 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1891 int offset
= offset_in_page(gpa
);
1894 while ((seg
= next_segment(len
, offset
)) != 0) {
1895 ret
= kvm_write_guest_page(kvm
, gfn
, data
, offset
, seg
);
1905 EXPORT_SYMBOL_GPL(kvm_write_guest
);
1907 int kvm_vcpu_write_guest(struct kvm_vcpu
*vcpu
, gpa_t gpa
, const void *data
,
1910 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
1912 int offset
= offset_in_page(gpa
);
1915 while ((seg
= next_segment(len
, offset
)) != 0) {
1916 ret
= kvm_vcpu_write_guest_page(vcpu
, gfn
, data
, offset
, seg
);
1926 EXPORT_SYMBOL_GPL(kvm_vcpu_write_guest
);
1928 int kvm_gfn_to_hva_cache_init(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1929 gpa_t gpa
, unsigned long len
)
1931 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1932 int offset
= offset_in_page(gpa
);
1933 gfn_t start_gfn
= gpa
>> PAGE_SHIFT
;
1934 gfn_t end_gfn
= (gpa
+ len
- 1) >> PAGE_SHIFT
;
1935 gfn_t nr_pages_needed
= end_gfn
- start_gfn
+ 1;
1936 gfn_t nr_pages_avail
;
1939 ghc
->generation
= slots
->generation
;
1941 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1942 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
, NULL
);
1943 if (!kvm_is_error_hva(ghc
->hva
) && nr_pages_needed
<= 1) {
1947 * If the requested region crosses two memslots, we still
1948 * verify that the entire region is valid here.
1950 while (start_gfn
<= end_gfn
) {
1951 ghc
->memslot
= gfn_to_memslot(kvm
, start_gfn
);
1952 ghc
->hva
= gfn_to_hva_many(ghc
->memslot
, start_gfn
,
1954 if (kvm_is_error_hva(ghc
->hva
))
1956 start_gfn
+= nr_pages_avail
;
1958 /* Use the slow path for cross page reads and writes. */
1959 ghc
->memslot
= NULL
;
1963 EXPORT_SYMBOL_GPL(kvm_gfn_to_hva_cache_init
);
1965 int kvm_write_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1966 void *data
, unsigned long len
)
1968 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1971 BUG_ON(len
> ghc
->len
);
1973 if (slots
->generation
!= ghc
->generation
)
1974 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
1976 if (unlikely(!ghc
->memslot
))
1977 return kvm_write_guest(kvm
, ghc
->gpa
, data
, len
);
1979 if (kvm_is_error_hva(ghc
->hva
))
1982 r
= __copy_to_user((void __user
*)ghc
->hva
, data
, len
);
1985 mark_page_dirty_in_slot(ghc
->memslot
, ghc
->gpa
>> PAGE_SHIFT
);
1989 EXPORT_SYMBOL_GPL(kvm_write_guest_cached
);
1991 int kvm_read_guest_cached(struct kvm
*kvm
, struct gfn_to_hva_cache
*ghc
,
1992 void *data
, unsigned long len
)
1994 struct kvm_memslots
*slots
= kvm_memslots(kvm
);
1997 BUG_ON(len
> ghc
->len
);
1999 if (slots
->generation
!= ghc
->generation
)
2000 kvm_gfn_to_hva_cache_init(kvm
, ghc
, ghc
->gpa
, ghc
->len
);
2002 if (unlikely(!ghc
->memslot
))
2003 return kvm_read_guest(kvm
, ghc
->gpa
, data
, len
);
2005 if (kvm_is_error_hva(ghc
->hva
))
2008 r
= __copy_from_user(data
, (void __user
*)ghc
->hva
, len
);
2014 EXPORT_SYMBOL_GPL(kvm_read_guest_cached
);
2016 int kvm_clear_guest_page(struct kvm
*kvm
, gfn_t gfn
, int offset
, int len
)
2018 const void *zero_page
= (const void *) __va(page_to_phys(ZERO_PAGE(0)));
2020 return kvm_write_guest_page(kvm
, gfn
, zero_page
, offset
, len
);
2022 EXPORT_SYMBOL_GPL(kvm_clear_guest_page
);
2024 int kvm_clear_guest(struct kvm
*kvm
, gpa_t gpa
, unsigned long len
)
2026 gfn_t gfn
= gpa
>> PAGE_SHIFT
;
2028 int offset
= offset_in_page(gpa
);
2031 while ((seg
= next_segment(len
, offset
)) != 0) {
2032 ret
= kvm_clear_guest_page(kvm
, gfn
, offset
, seg
);
2041 EXPORT_SYMBOL_GPL(kvm_clear_guest
);
2043 static void mark_page_dirty_in_slot(struct kvm_memory_slot
*memslot
,
2046 if (memslot
&& memslot
->dirty_bitmap
) {
2047 unsigned long rel_gfn
= gfn
- memslot
->base_gfn
;
2049 set_bit_le(rel_gfn
, memslot
->dirty_bitmap
);
2053 void mark_page_dirty(struct kvm
*kvm
, gfn_t gfn
)
2055 struct kvm_memory_slot
*memslot
;
2057 memslot
= gfn_to_memslot(kvm
, gfn
);
2058 mark_page_dirty_in_slot(memslot
, gfn
);
2060 EXPORT_SYMBOL_GPL(mark_page_dirty
);
2062 void kvm_vcpu_mark_page_dirty(struct kvm_vcpu
*vcpu
, gfn_t gfn
)
2064 struct kvm_memory_slot
*memslot
;
2066 memslot
= kvm_vcpu_gfn_to_memslot(vcpu
, gfn
);
2067 mark_page_dirty_in_slot(memslot
, gfn
);
2069 EXPORT_SYMBOL_GPL(kvm_vcpu_mark_page_dirty
);
2071 static void grow_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2073 unsigned int old
, val
, grow
;
2075 old
= val
= vcpu
->halt_poll_ns
;
2076 grow
= READ_ONCE(halt_poll_ns_grow
);
2078 if (val
== 0 && grow
)
2083 if (val
> halt_poll_ns
)
2086 vcpu
->halt_poll_ns
= val
;
2087 trace_kvm_halt_poll_ns_grow(vcpu
->vcpu_id
, val
, old
);
2090 static void shrink_halt_poll_ns(struct kvm_vcpu
*vcpu
)
2092 unsigned int old
, val
, shrink
;
2094 old
= val
= vcpu
->halt_poll_ns
;
2095 shrink
= READ_ONCE(halt_poll_ns_shrink
);
2101 vcpu
->halt_poll_ns
= val
;
2102 trace_kvm_halt_poll_ns_shrink(vcpu
->vcpu_id
, val
, old
);
2105 static int kvm_vcpu_check_block(struct kvm_vcpu
*vcpu
)
2107 if (kvm_arch_vcpu_runnable(vcpu
)) {
2108 kvm_make_request(KVM_REQ_UNHALT
, vcpu
);
2111 if (kvm_cpu_has_pending_timer(vcpu
))
2113 if (signal_pending(current
))
2120 * The vCPU has executed a HLT instruction with in-kernel mode enabled.
2122 void kvm_vcpu_block(struct kvm_vcpu
*vcpu
)
2125 DECLARE_SWAITQUEUE(wait
);
2126 bool waited
= false;
2129 start
= cur
= ktime_get();
2130 if (vcpu
->halt_poll_ns
) {
2131 ktime_t stop
= ktime_add_ns(ktime_get(), vcpu
->halt_poll_ns
);
2133 ++vcpu
->stat
.halt_attempted_poll
;
2136 * This sets KVM_REQ_UNHALT if an interrupt
2139 if (kvm_vcpu_check_block(vcpu
) < 0) {
2140 ++vcpu
->stat
.halt_successful_poll
;
2141 if (!vcpu_valid_wakeup(vcpu
))
2142 ++vcpu
->stat
.halt_poll_invalid
;
2146 } while (single_task_running() && ktime_before(cur
, stop
));
2149 kvm_arch_vcpu_blocking(vcpu
);
2152 prepare_to_swait(&vcpu
->wq
, &wait
, TASK_INTERRUPTIBLE
);
2154 if (kvm_vcpu_check_block(vcpu
) < 0)
2161 finish_swait(&vcpu
->wq
, &wait
);
2164 kvm_arch_vcpu_unblocking(vcpu
);
2166 block_ns
= ktime_to_ns(cur
) - ktime_to_ns(start
);
2168 if (!vcpu_valid_wakeup(vcpu
))
2169 shrink_halt_poll_ns(vcpu
);
2170 else if (halt_poll_ns
) {
2171 if (block_ns
<= vcpu
->halt_poll_ns
)
2173 /* we had a long block, shrink polling */
2174 else if (vcpu
->halt_poll_ns
&& block_ns
> halt_poll_ns
)
2175 shrink_halt_poll_ns(vcpu
);
2176 /* we had a short halt and our poll time is too small */
2177 else if (vcpu
->halt_poll_ns
< halt_poll_ns
&&
2178 block_ns
< halt_poll_ns
)
2179 grow_halt_poll_ns(vcpu
);
2181 vcpu
->halt_poll_ns
= 0;
2183 trace_kvm_vcpu_wakeup(block_ns
, waited
, vcpu_valid_wakeup(vcpu
));
2184 kvm_arch_vcpu_block_finish(vcpu
);
2186 EXPORT_SYMBOL_GPL(kvm_vcpu_block
);
2189 void kvm_vcpu_wake_up(struct kvm_vcpu
*vcpu
)
2191 struct swait_queue_head
*wqp
;
2193 wqp
= kvm_arch_vcpu_wq(vcpu
);
2194 if (swait_active(wqp
)) {
2196 ++vcpu
->stat
.halt_wakeup
;
2200 EXPORT_SYMBOL_GPL(kvm_vcpu_wake_up
);
2203 * Kick a sleeping VCPU, or a guest VCPU in guest mode, into host kernel mode.
2205 void kvm_vcpu_kick(struct kvm_vcpu
*vcpu
)
2208 int cpu
= vcpu
->cpu
;
2210 kvm_vcpu_wake_up(vcpu
);
2212 if (cpu
!= me
&& (unsigned)cpu
< nr_cpu_ids
&& cpu_online(cpu
))
2213 if (kvm_arch_vcpu_should_kick(vcpu
))
2214 smp_send_reschedule(cpu
);
2217 EXPORT_SYMBOL_GPL(kvm_vcpu_kick
);
2218 #endif /* !CONFIG_S390 */
2220 int kvm_vcpu_yield_to(struct kvm_vcpu
*target
)
2223 struct task_struct
*task
= NULL
;
2227 pid
= rcu_dereference(target
->pid
);
2229 task
= get_pid_task(pid
, PIDTYPE_PID
);
2233 ret
= yield_to(task
, 1);
2234 put_task_struct(task
);
2238 EXPORT_SYMBOL_GPL(kvm_vcpu_yield_to
);
2241 * Helper that checks whether a VCPU is eligible for directed yield.
2242 * Most eligible candidate to yield is decided by following heuristics:
2244 * (a) VCPU which has not done pl-exit or cpu relax intercepted recently
2245 * (preempted lock holder), indicated by @in_spin_loop.
2246 * Set at the beiginning and cleared at the end of interception/PLE handler.
2248 * (b) VCPU which has done pl-exit/ cpu relax intercepted but did not get
2249 * chance last time (mostly it has become eligible now since we have probably
2250 * yielded to lockholder in last iteration. This is done by toggling
2251 * @dy_eligible each time a VCPU checked for eligibility.)
2253 * Yielding to a recently pl-exited/cpu relax intercepted VCPU before yielding
2254 * to preempted lock-holder could result in wrong VCPU selection and CPU
2255 * burning. Giving priority for a potential lock-holder increases lock
2258 * Since algorithm is based on heuristics, accessing another VCPU data without
2259 * locking does not harm. It may result in trying to yield to same VCPU, fail
2260 * and continue with next VCPU and so on.
2262 static bool kvm_vcpu_eligible_for_directed_yield(struct kvm_vcpu
*vcpu
)
2264 #ifdef CONFIG_HAVE_KVM_CPU_RELAX_INTERCEPT
2267 eligible
= !vcpu
->spin_loop
.in_spin_loop
||
2268 vcpu
->spin_loop
.dy_eligible
;
2270 if (vcpu
->spin_loop
.in_spin_loop
)
2271 kvm_vcpu_set_dy_eligible(vcpu
, !vcpu
->spin_loop
.dy_eligible
);
2279 void kvm_vcpu_on_spin(struct kvm_vcpu
*me
)
2281 struct kvm
*kvm
= me
->kvm
;
2282 struct kvm_vcpu
*vcpu
;
2283 int last_boosted_vcpu
= me
->kvm
->last_boosted_vcpu
;
2289 kvm_vcpu_set_in_spin_loop(me
, true);
2291 * We boost the priority of a VCPU that is runnable but not
2292 * currently running, because it got preempted by something
2293 * else and called schedule in __vcpu_run. Hopefully that
2294 * VCPU is holding the lock that we need and will release it.
2295 * We approximate round-robin by starting at the last boosted VCPU.
2297 for (pass
= 0; pass
< 2 && !yielded
&& try; pass
++) {
2298 kvm_for_each_vcpu(i
, vcpu
, kvm
) {
2299 if (!pass
&& i
<= last_boosted_vcpu
) {
2300 i
= last_boosted_vcpu
;
2302 } else if (pass
&& i
> last_boosted_vcpu
)
2304 if (!ACCESS_ONCE(vcpu
->preempted
))
2308 if (swait_active(&vcpu
->wq
) && !kvm_arch_vcpu_runnable(vcpu
))
2310 if (!kvm_vcpu_eligible_for_directed_yield(vcpu
))
2313 yielded
= kvm_vcpu_yield_to(vcpu
);
2315 kvm
->last_boosted_vcpu
= i
;
2317 } else if (yielded
< 0) {
2324 kvm_vcpu_set_in_spin_loop(me
, false);
2326 /* Ensure vcpu is not eligible during next spinloop */
2327 kvm_vcpu_set_dy_eligible(me
, false);
2329 EXPORT_SYMBOL_GPL(kvm_vcpu_on_spin
);
2331 static int kvm_vcpu_fault(struct vm_area_struct
*vma
, struct vm_fault
*vmf
)
2333 struct kvm_vcpu
*vcpu
= vma
->vm_file
->private_data
;
2336 if (vmf
->pgoff
== 0)
2337 page
= virt_to_page(vcpu
->run
);
2339 else if (vmf
->pgoff
== KVM_PIO_PAGE_OFFSET
)
2340 page
= virt_to_page(vcpu
->arch
.pio_data
);
2342 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2343 else if (vmf
->pgoff
== KVM_COALESCED_MMIO_PAGE_OFFSET
)
2344 page
= virt_to_page(vcpu
->kvm
->coalesced_mmio_ring
);
2347 return kvm_arch_vcpu_fault(vcpu
, vmf
);
2353 static const struct vm_operations_struct kvm_vcpu_vm_ops
= {
2354 .fault
= kvm_vcpu_fault
,
2357 static int kvm_vcpu_mmap(struct file
*file
, struct vm_area_struct
*vma
)
2359 vma
->vm_ops
= &kvm_vcpu_vm_ops
;
2363 static int kvm_vcpu_release(struct inode
*inode
, struct file
*filp
)
2365 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2367 kvm_put_kvm(vcpu
->kvm
);
2371 static struct file_operations kvm_vcpu_fops
= {
2372 .release
= kvm_vcpu_release
,
2373 .unlocked_ioctl
= kvm_vcpu_ioctl
,
2374 #ifdef CONFIG_KVM_COMPAT
2375 .compat_ioctl
= kvm_vcpu_compat_ioctl
,
2377 .mmap
= kvm_vcpu_mmap
,
2378 .llseek
= noop_llseek
,
2382 * Allocates an inode for the vcpu.
2384 static int create_vcpu_fd(struct kvm_vcpu
*vcpu
)
2386 return anon_inode_getfd("kvm-vcpu", &kvm_vcpu_fops
, vcpu
, O_RDWR
| O_CLOEXEC
);
2390 * Creates some virtual cpus. Good luck creating more than one.
2392 static int kvm_vm_ioctl_create_vcpu(struct kvm
*kvm
, u32 id
)
2395 struct kvm_vcpu
*vcpu
;
2397 if (id
>= KVM_MAX_VCPU_ID
)
2400 mutex_lock(&kvm
->lock
);
2401 if (kvm
->created_vcpus
== KVM_MAX_VCPUS
) {
2402 mutex_unlock(&kvm
->lock
);
2406 kvm
->created_vcpus
++;
2407 mutex_unlock(&kvm
->lock
);
2409 vcpu
= kvm_arch_vcpu_create(kvm
, id
);
2412 goto vcpu_decrement
;
2415 preempt_notifier_init(&vcpu
->preempt_notifier
, &kvm_preempt_ops
);
2417 r
= kvm_arch_vcpu_setup(vcpu
);
2421 mutex_lock(&kvm
->lock
);
2422 if (kvm_get_vcpu_by_id(kvm
, id
)) {
2424 goto unlock_vcpu_destroy
;
2427 BUG_ON(kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)]);
2429 /* Now it's all set up, let userspace reach it */
2431 r
= create_vcpu_fd(vcpu
);
2434 goto unlock_vcpu_destroy
;
2437 kvm
->vcpus
[atomic_read(&kvm
->online_vcpus
)] = vcpu
;
2440 * Pairs with smp_rmb() in kvm_get_vcpu. Write kvm->vcpus
2441 * before kvm->online_vcpu's incremented value.
2444 atomic_inc(&kvm
->online_vcpus
);
2446 mutex_unlock(&kvm
->lock
);
2447 kvm_arch_vcpu_postcreate(vcpu
);
2450 unlock_vcpu_destroy
:
2451 mutex_unlock(&kvm
->lock
);
2453 kvm_arch_vcpu_destroy(vcpu
);
2455 mutex_lock(&kvm
->lock
);
2456 kvm
->created_vcpus
--;
2457 mutex_unlock(&kvm
->lock
);
2461 static int kvm_vcpu_ioctl_set_sigmask(struct kvm_vcpu
*vcpu
, sigset_t
*sigset
)
2464 sigdelsetmask(sigset
, sigmask(SIGKILL
)|sigmask(SIGSTOP
));
2465 vcpu
->sigset_active
= 1;
2466 vcpu
->sigset
= *sigset
;
2468 vcpu
->sigset_active
= 0;
2472 static long kvm_vcpu_ioctl(struct file
*filp
,
2473 unsigned int ioctl
, unsigned long arg
)
2475 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2476 void __user
*argp
= (void __user
*)arg
;
2478 struct kvm_fpu
*fpu
= NULL
;
2479 struct kvm_sregs
*kvm_sregs
= NULL
;
2481 if (vcpu
->kvm
->mm
!= current
->mm
)
2484 if (unlikely(_IOC_TYPE(ioctl
) != KVMIO
))
2487 #if defined(CONFIG_S390) || defined(CONFIG_PPC) || defined(CONFIG_MIPS)
2489 * Special cases: vcpu ioctls that are asynchronous to vcpu execution,
2490 * so vcpu_load() would break it.
2492 if (ioctl
== KVM_S390_INTERRUPT
|| ioctl
== KVM_S390_IRQ
|| ioctl
== KVM_INTERRUPT
)
2493 return kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2497 r
= vcpu_load(vcpu
);
2505 if (unlikely(vcpu
->pid
!= current
->pids
[PIDTYPE_PID
].pid
)) {
2506 /* The thread running this VCPU changed. */
2507 struct pid
*oldpid
= vcpu
->pid
;
2508 struct pid
*newpid
= get_task_pid(current
, PIDTYPE_PID
);
2510 rcu_assign_pointer(vcpu
->pid
, newpid
);
2515 r
= kvm_arch_vcpu_ioctl_run(vcpu
, vcpu
->run
);
2516 trace_kvm_userspace_exit(vcpu
->run
->exit_reason
, r
);
2518 case KVM_GET_REGS
: {
2519 struct kvm_regs
*kvm_regs
;
2522 kvm_regs
= kzalloc(sizeof(struct kvm_regs
), GFP_KERNEL
);
2525 r
= kvm_arch_vcpu_ioctl_get_regs(vcpu
, kvm_regs
);
2529 if (copy_to_user(argp
, kvm_regs
, sizeof(struct kvm_regs
)))
2536 case KVM_SET_REGS
: {
2537 struct kvm_regs
*kvm_regs
;
2540 kvm_regs
= memdup_user(argp
, sizeof(*kvm_regs
));
2541 if (IS_ERR(kvm_regs
)) {
2542 r
= PTR_ERR(kvm_regs
);
2545 r
= kvm_arch_vcpu_ioctl_set_regs(vcpu
, kvm_regs
);
2549 case KVM_GET_SREGS
: {
2550 kvm_sregs
= kzalloc(sizeof(struct kvm_sregs
), GFP_KERNEL
);
2554 r
= kvm_arch_vcpu_ioctl_get_sregs(vcpu
, kvm_sregs
);
2558 if (copy_to_user(argp
, kvm_sregs
, sizeof(struct kvm_sregs
)))
2563 case KVM_SET_SREGS
: {
2564 kvm_sregs
= memdup_user(argp
, sizeof(*kvm_sregs
));
2565 if (IS_ERR(kvm_sregs
)) {
2566 r
= PTR_ERR(kvm_sregs
);
2570 r
= kvm_arch_vcpu_ioctl_set_sregs(vcpu
, kvm_sregs
);
2573 case KVM_GET_MP_STATE
: {
2574 struct kvm_mp_state mp_state
;
2576 r
= kvm_arch_vcpu_ioctl_get_mpstate(vcpu
, &mp_state
);
2580 if (copy_to_user(argp
, &mp_state
, sizeof(mp_state
)))
2585 case KVM_SET_MP_STATE
: {
2586 struct kvm_mp_state mp_state
;
2589 if (copy_from_user(&mp_state
, argp
, sizeof(mp_state
)))
2591 r
= kvm_arch_vcpu_ioctl_set_mpstate(vcpu
, &mp_state
);
2594 case KVM_TRANSLATE
: {
2595 struct kvm_translation tr
;
2598 if (copy_from_user(&tr
, argp
, sizeof(tr
)))
2600 r
= kvm_arch_vcpu_ioctl_translate(vcpu
, &tr
);
2604 if (copy_to_user(argp
, &tr
, sizeof(tr
)))
2609 case KVM_SET_GUEST_DEBUG
: {
2610 struct kvm_guest_debug dbg
;
2613 if (copy_from_user(&dbg
, argp
, sizeof(dbg
)))
2615 r
= kvm_arch_vcpu_ioctl_set_guest_debug(vcpu
, &dbg
);
2618 case KVM_SET_SIGNAL_MASK
: {
2619 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2620 struct kvm_signal_mask kvm_sigmask
;
2621 sigset_t sigset
, *p
;
2626 if (copy_from_user(&kvm_sigmask
, argp
,
2627 sizeof(kvm_sigmask
)))
2630 if (kvm_sigmask
.len
!= sizeof(sigset
))
2633 if (copy_from_user(&sigset
, sigmask_arg
->sigset
,
2638 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, p
);
2642 fpu
= kzalloc(sizeof(struct kvm_fpu
), GFP_KERNEL
);
2646 r
= kvm_arch_vcpu_ioctl_get_fpu(vcpu
, fpu
);
2650 if (copy_to_user(argp
, fpu
, sizeof(struct kvm_fpu
)))
2656 fpu
= memdup_user(argp
, sizeof(*fpu
));
2662 r
= kvm_arch_vcpu_ioctl_set_fpu(vcpu
, fpu
);
2666 r
= kvm_arch_vcpu_ioctl(filp
, ioctl
, arg
);
2675 #ifdef CONFIG_KVM_COMPAT
2676 static long kvm_vcpu_compat_ioctl(struct file
*filp
,
2677 unsigned int ioctl
, unsigned long arg
)
2679 struct kvm_vcpu
*vcpu
= filp
->private_data
;
2680 void __user
*argp
= compat_ptr(arg
);
2683 if (vcpu
->kvm
->mm
!= current
->mm
)
2687 case KVM_SET_SIGNAL_MASK
: {
2688 struct kvm_signal_mask __user
*sigmask_arg
= argp
;
2689 struct kvm_signal_mask kvm_sigmask
;
2690 compat_sigset_t csigset
;
2695 if (copy_from_user(&kvm_sigmask
, argp
,
2696 sizeof(kvm_sigmask
)))
2699 if (kvm_sigmask
.len
!= sizeof(csigset
))
2702 if (copy_from_user(&csigset
, sigmask_arg
->sigset
,
2705 sigset_from_compat(&sigset
, &csigset
);
2706 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, &sigset
);
2708 r
= kvm_vcpu_ioctl_set_sigmask(vcpu
, NULL
);
2712 r
= kvm_vcpu_ioctl(filp
, ioctl
, arg
);
2720 static int kvm_device_ioctl_attr(struct kvm_device
*dev
,
2721 int (*accessor
)(struct kvm_device
*dev
,
2722 struct kvm_device_attr
*attr
),
2725 struct kvm_device_attr attr
;
2730 if (copy_from_user(&attr
, (void __user
*)arg
, sizeof(attr
)))
2733 return accessor(dev
, &attr
);
2736 static long kvm_device_ioctl(struct file
*filp
, unsigned int ioctl
,
2739 struct kvm_device
*dev
= filp
->private_data
;
2742 case KVM_SET_DEVICE_ATTR
:
2743 return kvm_device_ioctl_attr(dev
, dev
->ops
->set_attr
, arg
);
2744 case KVM_GET_DEVICE_ATTR
:
2745 return kvm_device_ioctl_attr(dev
, dev
->ops
->get_attr
, arg
);
2746 case KVM_HAS_DEVICE_ATTR
:
2747 return kvm_device_ioctl_attr(dev
, dev
->ops
->has_attr
, arg
);
2749 if (dev
->ops
->ioctl
)
2750 return dev
->ops
->ioctl(dev
, ioctl
, arg
);
2756 static int kvm_device_release(struct inode
*inode
, struct file
*filp
)
2758 struct kvm_device
*dev
= filp
->private_data
;
2759 struct kvm
*kvm
= dev
->kvm
;
2765 static const struct file_operations kvm_device_fops
= {
2766 .unlocked_ioctl
= kvm_device_ioctl
,
2767 #ifdef CONFIG_KVM_COMPAT
2768 .compat_ioctl
= kvm_device_ioctl
,
2770 .release
= kvm_device_release
,
2773 struct kvm_device
*kvm_device_from_filp(struct file
*filp
)
2775 if (filp
->f_op
!= &kvm_device_fops
)
2778 return filp
->private_data
;
2781 static struct kvm_device_ops
*kvm_device_ops_table
[KVM_DEV_TYPE_MAX
] = {
2782 #ifdef CONFIG_KVM_MPIC
2783 [KVM_DEV_TYPE_FSL_MPIC_20
] = &kvm_mpic_ops
,
2784 [KVM_DEV_TYPE_FSL_MPIC_42
] = &kvm_mpic_ops
,
2787 #ifdef CONFIG_KVM_XICS
2788 [KVM_DEV_TYPE_XICS
] = &kvm_xics_ops
,
2792 int kvm_register_device_ops(struct kvm_device_ops
*ops
, u32 type
)
2794 if (type
>= ARRAY_SIZE(kvm_device_ops_table
))
2797 if (kvm_device_ops_table
[type
] != NULL
)
2800 kvm_device_ops_table
[type
] = ops
;
2804 void kvm_unregister_device_ops(u32 type
)
2806 if (kvm_device_ops_table
[type
] != NULL
)
2807 kvm_device_ops_table
[type
] = NULL
;
2810 static int kvm_ioctl_create_device(struct kvm
*kvm
,
2811 struct kvm_create_device
*cd
)
2813 struct kvm_device_ops
*ops
= NULL
;
2814 struct kvm_device
*dev
;
2815 bool test
= cd
->flags
& KVM_CREATE_DEVICE_TEST
;
2818 if (cd
->type
>= ARRAY_SIZE(kvm_device_ops_table
))
2821 ops
= kvm_device_ops_table
[cd
->type
];
2828 dev
= kzalloc(sizeof(*dev
), GFP_KERNEL
);
2835 ret
= ops
->create(dev
, cd
->type
);
2841 ret
= anon_inode_getfd(ops
->name
, &kvm_device_fops
, dev
, O_RDWR
| O_CLOEXEC
);
2847 list_add(&dev
->vm_node
, &kvm
->devices
);
2853 static long kvm_vm_ioctl_check_extension_generic(struct kvm
*kvm
, long arg
)
2856 case KVM_CAP_USER_MEMORY
:
2857 case KVM_CAP_DESTROY_MEMORY_REGION_WORKS
:
2858 case KVM_CAP_JOIN_MEMORY_REGIONS_WORKS
:
2859 case KVM_CAP_INTERNAL_ERROR_DATA
:
2860 #ifdef CONFIG_HAVE_KVM_MSI
2861 case KVM_CAP_SIGNAL_MSI
:
2863 #ifdef CONFIG_HAVE_KVM_IRQFD
2865 case KVM_CAP_IRQFD_RESAMPLE
:
2867 case KVM_CAP_IOEVENTFD_ANY_LENGTH
:
2868 case KVM_CAP_CHECK_EXTENSION_VM
:
2870 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2871 case KVM_CAP_IRQ_ROUTING
:
2872 return KVM_MAX_IRQ_ROUTES
;
2874 #if KVM_ADDRESS_SPACE_NUM > 1
2875 case KVM_CAP_MULTI_ADDRESS_SPACE
:
2876 return KVM_ADDRESS_SPACE_NUM
;
2878 case KVM_CAP_MAX_VCPU_ID
:
2879 return KVM_MAX_VCPU_ID
;
2883 return kvm_vm_ioctl_check_extension(kvm
, arg
);
2886 static long kvm_vm_ioctl(struct file
*filp
,
2887 unsigned int ioctl
, unsigned long arg
)
2889 struct kvm
*kvm
= filp
->private_data
;
2890 void __user
*argp
= (void __user
*)arg
;
2893 if (kvm
->mm
!= current
->mm
)
2896 case KVM_CREATE_VCPU
:
2897 r
= kvm_vm_ioctl_create_vcpu(kvm
, arg
);
2899 case KVM_SET_USER_MEMORY_REGION
: {
2900 struct kvm_userspace_memory_region kvm_userspace_mem
;
2903 if (copy_from_user(&kvm_userspace_mem
, argp
,
2904 sizeof(kvm_userspace_mem
)))
2907 r
= kvm_vm_ioctl_set_memory_region(kvm
, &kvm_userspace_mem
);
2910 case KVM_GET_DIRTY_LOG
: {
2911 struct kvm_dirty_log log
;
2914 if (copy_from_user(&log
, argp
, sizeof(log
)))
2916 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
2919 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
2920 case KVM_REGISTER_COALESCED_MMIO
: {
2921 struct kvm_coalesced_mmio_zone zone
;
2924 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2926 r
= kvm_vm_ioctl_register_coalesced_mmio(kvm
, &zone
);
2929 case KVM_UNREGISTER_COALESCED_MMIO
: {
2930 struct kvm_coalesced_mmio_zone zone
;
2933 if (copy_from_user(&zone
, argp
, sizeof(zone
)))
2935 r
= kvm_vm_ioctl_unregister_coalesced_mmio(kvm
, &zone
);
2940 struct kvm_irqfd data
;
2943 if (copy_from_user(&data
, argp
, sizeof(data
)))
2945 r
= kvm_irqfd(kvm
, &data
);
2948 case KVM_IOEVENTFD
: {
2949 struct kvm_ioeventfd data
;
2952 if (copy_from_user(&data
, argp
, sizeof(data
)))
2954 r
= kvm_ioeventfd(kvm
, &data
);
2957 #ifdef CONFIG_HAVE_KVM_MSI
2958 case KVM_SIGNAL_MSI
: {
2962 if (copy_from_user(&msi
, argp
, sizeof(msi
)))
2964 r
= kvm_send_userspace_msi(kvm
, &msi
);
2968 #ifdef __KVM_HAVE_IRQ_LINE
2969 case KVM_IRQ_LINE_STATUS
:
2970 case KVM_IRQ_LINE
: {
2971 struct kvm_irq_level irq_event
;
2974 if (copy_from_user(&irq_event
, argp
, sizeof(irq_event
)))
2977 r
= kvm_vm_ioctl_irq_line(kvm
, &irq_event
,
2978 ioctl
== KVM_IRQ_LINE_STATUS
);
2983 if (ioctl
== KVM_IRQ_LINE_STATUS
) {
2984 if (copy_to_user(argp
, &irq_event
, sizeof(irq_event
)))
2992 #ifdef CONFIG_HAVE_KVM_IRQ_ROUTING
2993 case KVM_SET_GSI_ROUTING
: {
2994 struct kvm_irq_routing routing
;
2995 struct kvm_irq_routing __user
*urouting
;
2996 struct kvm_irq_routing_entry
*entries
= NULL
;
2999 if (copy_from_user(&routing
, argp
, sizeof(routing
)))
3002 if (routing
.nr
> KVM_MAX_IRQ_ROUTES
)
3008 entries
= vmalloc(routing
.nr
* sizeof(*entries
));
3013 if (copy_from_user(entries
, urouting
->entries
,
3014 routing
.nr
* sizeof(*entries
)))
3015 goto out_free_irq_routing
;
3017 r
= kvm_set_irq_routing(kvm
, entries
, routing
.nr
,
3019 out_free_irq_routing
:
3023 #endif /* CONFIG_HAVE_KVM_IRQ_ROUTING */
3024 case KVM_CREATE_DEVICE
: {
3025 struct kvm_create_device cd
;
3028 if (copy_from_user(&cd
, argp
, sizeof(cd
)))
3031 r
= kvm_ioctl_create_device(kvm
, &cd
);
3036 if (copy_to_user(argp
, &cd
, sizeof(cd
)))
3042 case KVM_CHECK_EXTENSION
:
3043 r
= kvm_vm_ioctl_check_extension_generic(kvm
, arg
);
3046 r
= kvm_arch_vm_ioctl(filp
, ioctl
, arg
);
3052 #ifdef CONFIG_KVM_COMPAT
3053 struct compat_kvm_dirty_log
{
3057 compat_uptr_t dirty_bitmap
; /* one bit per page */
3062 static long kvm_vm_compat_ioctl(struct file
*filp
,
3063 unsigned int ioctl
, unsigned long arg
)
3065 struct kvm
*kvm
= filp
->private_data
;
3068 if (kvm
->mm
!= current
->mm
)
3071 case KVM_GET_DIRTY_LOG
: {
3072 struct compat_kvm_dirty_log compat_log
;
3073 struct kvm_dirty_log log
;
3076 if (copy_from_user(&compat_log
, (void __user
*)arg
,
3077 sizeof(compat_log
)))
3079 log
.slot
= compat_log
.slot
;
3080 log
.padding1
= compat_log
.padding1
;
3081 log
.padding2
= compat_log
.padding2
;
3082 log
.dirty_bitmap
= compat_ptr(compat_log
.dirty_bitmap
);
3084 r
= kvm_vm_ioctl_get_dirty_log(kvm
, &log
);
3088 r
= kvm_vm_ioctl(filp
, ioctl
, arg
);
3096 static struct file_operations kvm_vm_fops
= {
3097 .release
= kvm_vm_release
,
3098 .unlocked_ioctl
= kvm_vm_ioctl
,
3099 #ifdef CONFIG_KVM_COMPAT
3100 .compat_ioctl
= kvm_vm_compat_ioctl
,
3102 .llseek
= noop_llseek
,
3105 static int kvm_dev_ioctl_create_vm(unsigned long type
)
3111 kvm
= kvm_create_vm(type
);
3113 return PTR_ERR(kvm
);
3114 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3115 r
= kvm_coalesced_mmio_init(kvm
);
3121 r
= get_unused_fd_flags(O_CLOEXEC
);
3126 file
= anon_inode_getfile("kvm-vm", &kvm_vm_fops
, kvm
, O_RDWR
);
3130 return PTR_ERR(file
);
3133 if (kvm_create_vm_debugfs(kvm
, r
) < 0) {
3139 fd_install(r
, file
);
3143 static long kvm_dev_ioctl(struct file
*filp
,
3144 unsigned int ioctl
, unsigned long arg
)
3149 case KVM_GET_API_VERSION
:
3152 r
= KVM_API_VERSION
;
3155 r
= kvm_dev_ioctl_create_vm(arg
);
3157 case KVM_CHECK_EXTENSION
:
3158 r
= kvm_vm_ioctl_check_extension_generic(NULL
, arg
);
3160 case KVM_GET_VCPU_MMAP_SIZE
:
3163 r
= PAGE_SIZE
; /* struct kvm_run */
3165 r
+= PAGE_SIZE
; /* pio data page */
3167 #ifdef KVM_COALESCED_MMIO_PAGE_OFFSET
3168 r
+= PAGE_SIZE
; /* coalesced mmio ring page */
3171 case KVM_TRACE_ENABLE
:
3172 case KVM_TRACE_PAUSE
:
3173 case KVM_TRACE_DISABLE
:
3177 return kvm_arch_dev_ioctl(filp
, ioctl
, arg
);
3183 static struct file_operations kvm_chardev_ops
= {
3184 .unlocked_ioctl
= kvm_dev_ioctl
,
3185 .compat_ioctl
= kvm_dev_ioctl
,
3186 .llseek
= noop_llseek
,
3189 static struct miscdevice kvm_dev
= {
3195 static void hardware_enable_nolock(void *junk
)
3197 int cpu
= raw_smp_processor_id();
3200 if (cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3203 cpumask_set_cpu(cpu
, cpus_hardware_enabled
);
3205 r
= kvm_arch_hardware_enable();
3208 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3209 atomic_inc(&hardware_enable_failed
);
3210 pr_info("kvm: enabling virtualization on CPU%d failed\n", cpu
);
3214 static int kvm_starting_cpu(unsigned int cpu
)
3216 raw_spin_lock(&kvm_count_lock
);
3217 if (kvm_usage_count
)
3218 hardware_enable_nolock(NULL
);
3219 raw_spin_unlock(&kvm_count_lock
);
3223 static void hardware_disable_nolock(void *junk
)
3225 int cpu
= raw_smp_processor_id();
3227 if (!cpumask_test_cpu(cpu
, cpus_hardware_enabled
))
3229 cpumask_clear_cpu(cpu
, cpus_hardware_enabled
);
3230 kvm_arch_hardware_disable();
3233 static int kvm_dying_cpu(unsigned int cpu
)
3235 raw_spin_lock(&kvm_count_lock
);
3236 if (kvm_usage_count
)
3237 hardware_disable_nolock(NULL
);
3238 raw_spin_unlock(&kvm_count_lock
);
3242 static void hardware_disable_all_nolock(void)
3244 BUG_ON(!kvm_usage_count
);
3247 if (!kvm_usage_count
)
3248 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3251 static void hardware_disable_all(void)
3253 raw_spin_lock(&kvm_count_lock
);
3254 hardware_disable_all_nolock();
3255 raw_spin_unlock(&kvm_count_lock
);
3258 static int hardware_enable_all(void)
3262 raw_spin_lock(&kvm_count_lock
);
3265 if (kvm_usage_count
== 1) {
3266 atomic_set(&hardware_enable_failed
, 0);
3267 on_each_cpu(hardware_enable_nolock
, NULL
, 1);
3269 if (atomic_read(&hardware_enable_failed
)) {
3270 hardware_disable_all_nolock();
3275 raw_spin_unlock(&kvm_count_lock
);
3280 static int kvm_reboot(struct notifier_block
*notifier
, unsigned long val
,
3284 * Some (well, at least mine) BIOSes hang on reboot if
3287 * And Intel TXT required VMX off for all cpu when system shutdown.
3289 pr_info("kvm: exiting hardware virtualization\n");
3290 kvm_rebooting
= true;
3291 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3295 static struct notifier_block kvm_reboot_notifier
= {
3296 .notifier_call
= kvm_reboot
,
3300 static void kvm_io_bus_destroy(struct kvm_io_bus
*bus
)
3304 for (i
= 0; i
< bus
->dev_count
; i
++) {
3305 struct kvm_io_device
*pos
= bus
->range
[i
].dev
;
3307 kvm_iodevice_destructor(pos
);
3312 static inline int kvm_io_bus_cmp(const struct kvm_io_range
*r1
,
3313 const struct kvm_io_range
*r2
)
3315 gpa_t addr1
= r1
->addr
;
3316 gpa_t addr2
= r2
->addr
;
3321 /* If r2->len == 0, match the exact address. If r2->len != 0,
3322 * accept any overlapping write. Any order is acceptable for
3323 * overlapping ranges, because kvm_io_bus_get_first_dev ensures
3324 * we process all of them.
3337 static int kvm_io_bus_sort_cmp(const void *p1
, const void *p2
)
3339 return kvm_io_bus_cmp(p1
, p2
);
3342 static int kvm_io_bus_insert_dev(struct kvm_io_bus
*bus
, struct kvm_io_device
*dev
,
3343 gpa_t addr
, int len
)
3345 bus
->range
[bus
->dev_count
++] = (struct kvm_io_range
) {
3351 sort(bus
->range
, bus
->dev_count
, sizeof(struct kvm_io_range
),
3352 kvm_io_bus_sort_cmp
, NULL
);
3357 static int kvm_io_bus_get_first_dev(struct kvm_io_bus
*bus
,
3358 gpa_t addr
, int len
)
3360 struct kvm_io_range
*range
, key
;
3363 key
= (struct kvm_io_range
) {
3368 range
= bsearch(&key
, bus
->range
, bus
->dev_count
,
3369 sizeof(struct kvm_io_range
), kvm_io_bus_sort_cmp
);
3373 off
= range
- bus
->range
;
3375 while (off
> 0 && kvm_io_bus_cmp(&key
, &bus
->range
[off
-1]) == 0)
3381 static int __kvm_io_bus_write(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3382 struct kvm_io_range
*range
, const void *val
)
3386 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3390 while (idx
< bus
->dev_count
&&
3391 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3392 if (!kvm_iodevice_write(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3401 /* kvm_io_bus_write - called under kvm->slots_lock */
3402 int kvm_io_bus_write(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3403 int len
, const void *val
)
3405 struct kvm_io_bus
*bus
;
3406 struct kvm_io_range range
;
3409 range
= (struct kvm_io_range
) {
3414 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3415 r
= __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3416 return r
< 0 ? r
: 0;
3419 /* kvm_io_bus_write_cookie - called under kvm->slots_lock */
3420 int kvm_io_bus_write_cookie(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
,
3421 gpa_t addr
, int len
, const void *val
, long cookie
)
3423 struct kvm_io_bus
*bus
;
3424 struct kvm_io_range range
;
3426 range
= (struct kvm_io_range
) {
3431 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3433 /* First try the device referenced by cookie. */
3434 if ((cookie
>= 0) && (cookie
< bus
->dev_count
) &&
3435 (kvm_io_bus_cmp(&range
, &bus
->range
[cookie
]) == 0))
3436 if (!kvm_iodevice_write(vcpu
, bus
->range
[cookie
].dev
, addr
, len
,
3441 * cookie contained garbage; fall back to search and return the
3442 * correct cookie value.
3444 return __kvm_io_bus_write(vcpu
, bus
, &range
, val
);
3447 static int __kvm_io_bus_read(struct kvm_vcpu
*vcpu
, struct kvm_io_bus
*bus
,
3448 struct kvm_io_range
*range
, void *val
)
3452 idx
= kvm_io_bus_get_first_dev(bus
, range
->addr
, range
->len
);
3456 while (idx
< bus
->dev_count
&&
3457 kvm_io_bus_cmp(range
, &bus
->range
[idx
]) == 0) {
3458 if (!kvm_iodevice_read(vcpu
, bus
->range
[idx
].dev
, range
->addr
,
3466 EXPORT_SYMBOL_GPL(kvm_io_bus_write
);
3468 /* kvm_io_bus_read - called under kvm->slots_lock */
3469 int kvm_io_bus_read(struct kvm_vcpu
*vcpu
, enum kvm_bus bus_idx
, gpa_t addr
,
3472 struct kvm_io_bus
*bus
;
3473 struct kvm_io_range range
;
3476 range
= (struct kvm_io_range
) {
3481 bus
= srcu_dereference(vcpu
->kvm
->buses
[bus_idx
], &vcpu
->kvm
->srcu
);
3482 r
= __kvm_io_bus_read(vcpu
, bus
, &range
, val
);
3483 return r
< 0 ? r
: 0;
3487 /* Caller must hold slots_lock. */
3488 int kvm_io_bus_register_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
, gpa_t addr
,
3489 int len
, struct kvm_io_device
*dev
)
3491 struct kvm_io_bus
*new_bus
, *bus
;
3493 bus
= kvm
->buses
[bus_idx
];
3494 /* exclude ioeventfd which is limited by maximum fd */
3495 if (bus
->dev_count
- bus
->ioeventfd_count
> NR_IOBUS_DEVS
- 1)
3498 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
+ 1) *
3499 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3502 memcpy(new_bus
, bus
, sizeof(*bus
) + (bus
->dev_count
*
3503 sizeof(struct kvm_io_range
)));
3504 kvm_io_bus_insert_dev(new_bus
, dev
, addr
, len
);
3505 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3506 synchronize_srcu_expedited(&kvm
->srcu
);
3512 /* Caller must hold slots_lock. */
3513 int kvm_io_bus_unregister_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3514 struct kvm_io_device
*dev
)
3517 struct kvm_io_bus
*new_bus
, *bus
;
3519 bus
= kvm
->buses
[bus_idx
];
3521 for (i
= 0; i
< bus
->dev_count
; i
++)
3522 if (bus
->range
[i
].dev
== dev
) {
3530 new_bus
= kmalloc(sizeof(*bus
) + ((bus
->dev_count
- 1) *
3531 sizeof(struct kvm_io_range
)), GFP_KERNEL
);
3535 memcpy(new_bus
, bus
, sizeof(*bus
) + i
* sizeof(struct kvm_io_range
));
3536 new_bus
->dev_count
--;
3537 memcpy(new_bus
->range
+ i
, bus
->range
+ i
+ 1,
3538 (new_bus
->dev_count
- i
) * sizeof(struct kvm_io_range
));
3540 rcu_assign_pointer(kvm
->buses
[bus_idx
], new_bus
);
3541 synchronize_srcu_expedited(&kvm
->srcu
);
3546 struct kvm_io_device
*kvm_io_bus_get_dev(struct kvm
*kvm
, enum kvm_bus bus_idx
,
3549 struct kvm_io_bus
*bus
;
3550 int dev_idx
, srcu_idx
;
3551 struct kvm_io_device
*iodev
= NULL
;
3553 srcu_idx
= srcu_read_lock(&kvm
->srcu
);
3555 bus
= srcu_dereference(kvm
->buses
[bus_idx
], &kvm
->srcu
);
3557 dev_idx
= kvm_io_bus_get_first_dev(bus
, addr
, 1);
3561 iodev
= bus
->range
[dev_idx
].dev
;
3564 srcu_read_unlock(&kvm
->srcu
, srcu_idx
);
3568 EXPORT_SYMBOL_GPL(kvm_io_bus_get_dev
);
3570 static int kvm_debugfs_open(struct inode
*inode
, struct file
*file
,
3571 int (*get
)(void *, u64
*), int (*set
)(void *, u64
),
3574 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3577 /* The debugfs files are a reference to the kvm struct which
3578 * is still valid when kvm_destroy_vm is called.
3579 * To avoid the race between open and the removal of the debugfs
3580 * directory we test against the users count.
3582 if (!atomic_add_unless(&stat_data
->kvm
->users_count
, 1, 0))
3585 if (simple_attr_open(inode
, file
, get
, set
, fmt
)) {
3586 kvm_put_kvm(stat_data
->kvm
);
3593 static int kvm_debugfs_release(struct inode
*inode
, struct file
*file
)
3595 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)
3598 simple_attr_release(inode
, file
);
3599 kvm_put_kvm(stat_data
->kvm
);
3604 static int vm_stat_get_per_vm(void *data
, u64
*val
)
3606 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3608 *val
= *(u32
*)((void *)stat_data
->kvm
+ stat_data
->offset
);
3613 static int vm_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3615 __simple_attr_check_format("%llu\n", 0ull);
3616 return kvm_debugfs_open(inode
, file
, vm_stat_get_per_vm
,
3620 static const struct file_operations vm_stat_get_per_vm_fops
= {
3621 .owner
= THIS_MODULE
,
3622 .open
= vm_stat_get_per_vm_open
,
3623 .release
= kvm_debugfs_release
,
3624 .read
= simple_attr_read
,
3625 .write
= simple_attr_write
,
3626 .llseek
= generic_file_llseek
,
3629 static int vcpu_stat_get_per_vm(void *data
, u64
*val
)
3632 struct kvm_stat_data
*stat_data
= (struct kvm_stat_data
*)data
;
3633 struct kvm_vcpu
*vcpu
;
3637 kvm_for_each_vcpu(i
, vcpu
, stat_data
->kvm
)
3638 *val
+= *(u32
*)((void *)vcpu
+ stat_data
->offset
);
3643 static int vcpu_stat_get_per_vm_open(struct inode
*inode
, struct file
*file
)
3645 __simple_attr_check_format("%llu\n", 0ull);
3646 return kvm_debugfs_open(inode
, file
, vcpu_stat_get_per_vm
,
3650 static const struct file_operations vcpu_stat_get_per_vm_fops
= {
3651 .owner
= THIS_MODULE
,
3652 .open
= vcpu_stat_get_per_vm_open
,
3653 .release
= kvm_debugfs_release
,
3654 .read
= simple_attr_read
,
3655 .write
= simple_attr_write
,
3656 .llseek
= generic_file_llseek
,
3659 static const struct file_operations
*stat_fops_per_vm
[] = {
3660 [KVM_STAT_VCPU
] = &vcpu_stat_get_per_vm_fops
,
3661 [KVM_STAT_VM
] = &vm_stat_get_per_vm_fops
,
3664 static int vm_stat_get(void *_offset
, u64
*val
)
3666 unsigned offset
= (long)_offset
;
3668 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3672 spin_lock(&kvm_lock
);
3673 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3675 vm_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3678 spin_unlock(&kvm_lock
);
3682 DEFINE_SIMPLE_ATTRIBUTE(vm_stat_fops
, vm_stat_get
, NULL
, "%llu\n");
3684 static int vcpu_stat_get(void *_offset
, u64
*val
)
3686 unsigned offset
= (long)_offset
;
3688 struct kvm_stat_data stat_tmp
= {.offset
= offset
};
3692 spin_lock(&kvm_lock
);
3693 list_for_each_entry(kvm
, &vm_list
, vm_list
) {
3695 vcpu_stat_get_per_vm((void *)&stat_tmp
, &tmp_val
);
3698 spin_unlock(&kvm_lock
);
3702 DEFINE_SIMPLE_ATTRIBUTE(vcpu_stat_fops
, vcpu_stat_get
, NULL
, "%llu\n");
3704 static const struct file_operations
*stat_fops
[] = {
3705 [KVM_STAT_VCPU
] = &vcpu_stat_fops
,
3706 [KVM_STAT_VM
] = &vm_stat_fops
,
3709 static int kvm_init_debug(void)
3712 struct kvm_stats_debugfs_item
*p
;
3714 kvm_debugfs_dir
= debugfs_create_dir("kvm", NULL
);
3715 if (kvm_debugfs_dir
== NULL
)
3718 kvm_debugfs_num_entries
= 0;
3719 for (p
= debugfs_entries
; p
->name
; ++p
, kvm_debugfs_num_entries
++) {
3720 if (!debugfs_create_file(p
->name
, 0444, kvm_debugfs_dir
,
3721 (void *)(long)p
->offset
,
3722 stat_fops
[p
->kind
]))
3729 debugfs_remove_recursive(kvm_debugfs_dir
);
3734 static int kvm_suspend(void)
3736 if (kvm_usage_count
)
3737 hardware_disable_nolock(NULL
);
3741 static void kvm_resume(void)
3743 if (kvm_usage_count
) {
3744 WARN_ON(raw_spin_is_locked(&kvm_count_lock
));
3745 hardware_enable_nolock(NULL
);
3749 static struct syscore_ops kvm_syscore_ops
= {
3750 .suspend
= kvm_suspend
,
3751 .resume
= kvm_resume
,
3755 struct kvm_vcpu
*preempt_notifier_to_vcpu(struct preempt_notifier
*pn
)
3757 return container_of(pn
, struct kvm_vcpu
, preempt_notifier
);
3760 static void kvm_sched_in(struct preempt_notifier
*pn
, int cpu
)
3762 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3764 if (vcpu
->preempted
)
3765 vcpu
->preempted
= false;
3767 kvm_arch_sched_in(vcpu
, cpu
);
3769 kvm_arch_vcpu_load(vcpu
, cpu
);
3772 static void kvm_sched_out(struct preempt_notifier
*pn
,
3773 struct task_struct
*next
)
3775 struct kvm_vcpu
*vcpu
= preempt_notifier_to_vcpu(pn
);
3777 if (current
->state
== TASK_RUNNING
)
3778 vcpu
->preempted
= true;
3779 kvm_arch_vcpu_put(vcpu
);
3782 int kvm_init(void *opaque
, unsigned vcpu_size
, unsigned vcpu_align
,
3783 struct module
*module
)
3788 r
= kvm_arch_init(opaque
);
3793 * kvm_arch_init makes sure there's at most one caller
3794 * for architectures that support multiple implementations,
3795 * like intel and amd on x86.
3796 * kvm_arch_init must be called before kvm_irqfd_init to avoid creating
3797 * conflicts in case kvm is already setup for another implementation.
3799 r
= kvm_irqfd_init();
3803 if (!zalloc_cpumask_var(&cpus_hardware_enabled
, GFP_KERNEL
)) {
3808 r
= kvm_arch_hardware_setup();
3812 for_each_online_cpu(cpu
) {
3813 smp_call_function_single(cpu
,
3814 kvm_arch_check_processor_compat
,
3820 r
= cpuhp_setup_state_nocalls(CPUHP_AP_KVM_STARTING
, "AP_KVM_STARTING",
3821 kvm_starting_cpu
, kvm_dying_cpu
);
3824 register_reboot_notifier(&kvm_reboot_notifier
);
3826 /* A kmem cache lets us meet the alignment requirements of fx_save. */
3828 vcpu_align
= __alignof__(struct kvm_vcpu
);
3829 kvm_vcpu_cache
= kmem_cache_create("kvm_vcpu", vcpu_size
, vcpu_align
,
3831 if (!kvm_vcpu_cache
) {
3836 r
= kvm_async_pf_init();
3840 kvm_chardev_ops
.owner
= module
;
3841 kvm_vm_fops
.owner
= module
;
3842 kvm_vcpu_fops
.owner
= module
;
3844 r
= misc_register(&kvm_dev
);
3846 pr_err("kvm: misc device register failed\n");
3850 register_syscore_ops(&kvm_syscore_ops
);
3852 kvm_preempt_ops
.sched_in
= kvm_sched_in
;
3853 kvm_preempt_ops
.sched_out
= kvm_sched_out
;
3855 r
= kvm_init_debug();
3857 pr_err("kvm: create debugfs files failed\n");
3861 r
= kvm_vfio_ops_init();
3867 unregister_syscore_ops(&kvm_syscore_ops
);
3868 misc_deregister(&kvm_dev
);
3870 kvm_async_pf_deinit();
3872 kmem_cache_destroy(kvm_vcpu_cache
);
3874 unregister_reboot_notifier(&kvm_reboot_notifier
);
3875 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
3878 kvm_arch_hardware_unsetup();
3880 free_cpumask_var(cpus_hardware_enabled
);
3888 EXPORT_SYMBOL_GPL(kvm_init
);
3892 debugfs_remove_recursive(kvm_debugfs_dir
);
3893 misc_deregister(&kvm_dev
);
3894 kmem_cache_destroy(kvm_vcpu_cache
);
3895 kvm_async_pf_deinit();
3896 unregister_syscore_ops(&kvm_syscore_ops
);
3897 unregister_reboot_notifier(&kvm_reboot_notifier
);
3898 cpuhp_remove_state_nocalls(CPUHP_AP_KVM_STARTING
);
3899 on_each_cpu(hardware_disable_nolock
, NULL
, 1);
3900 kvm_arch_hardware_unsetup();
3903 free_cpumask_var(cpus_hardware_enabled
);
3904 kvm_vfio_ops_exit();
3906 EXPORT_SYMBOL_GPL(kvm_exit
);